The Role of El Niño in Driving Drought Conditions over the Last 2000 Years in Thailand

: Irregular climate events frequently occur in Southeast Asia due to the numerous climate patterns combining. Thailand interactions, and consequently major hydrological events, such as Proxy speleothem records, lake sediment sequences and tree ring chronologies were used to reconstruct paleo drought conditions. These trends were compared with modelled and historic El Niño Southern Oscillation (ENSO) data to assess if the ENSO climate phenomena is causing droughts in Thailand. Drought periods were found to occur both during El Niño events and ENSO neutral conditions. This indicates droughts are not a product of one climate pattern, but likely the result of numerous patterns interacting. There is uncertainty regarding how climate patterns will evolve under climate change, but changes in amplitude and variability could potentially lead to more frequent and wider reaching hydrological disasters. It is vital that policies are implemented to cope with the resulting social and economic repercussions, including diversiﬁcation of crops and reorganisation of water consumption behaviour in Thailand.


Introduction
Anomalous climatic behaviour associated with the alternating wet and dry East Asian monsoon climate can cause devastating hydrological events across South East Asia [1]. A large effort has been focused on both understanding and forecasting East Asian monsoon variability [2,3], with research indicating changeability is due to a combination of localised factors, including sea surface temperature, snow cover and soil moisture [4], alongside global climate patterns such as the tropospheric biennial oscillation [5] and the El Niño Southern Oscillation (ENSO) [6,7]. The El Niño Southern Oscillation warm event is characterised by the sudden appearance of irregularly warm surface water in the tropical central and eastern Pacific Ocean off the coast of Peru [8]. There is a general relationship between the occurrence of El Niño events and deficiency in rainfall, in some cases causing agricultural drought conditions [9,10]. Below normal summer monsoon levels have been recorded during El Niño years, for the Indian [11], and East Asian monsoons [1]. An association between drought disasters and ENSO warm events in Asia and Africa has been identified [12], with the rate of people affected by drought disasters increasing globally 183% during El Niño years [13].
Thailand sits at the confluence of many different climatic influences. The climate of Thailand is tropical, with three distinct seasons: hot March to May, rainy season May to October and drier and cooler from November to February. The average annual climate for 1985 to 2015 is depicted in Figure 1. The rainy season is driven by the southwest monsoon, and the cooler season is a product from the prevailing north east monsoon [14]. It is strongly affected by the Asian monsoon systems [15,16] and is exposed to El Niño driven climatic anomalies. It also experiences the remnants of northwest Pacific tropical cyclones [17], and the fluctuating climatic behaviour is further modulated by the shifting of the ascending Walker circulation and the influence of the Intertropical Convergence Zone (ITCZ) [18]. 2010 was a unique year to showcase this climate variability. El Niño conditions triggered unusually hot weather and a rainfall deficiency in January by an early withdrawal of monsoon season by November 2009, evolving into national scale drought by March 2010 [19]. In July, the start of the monsoon season coupled with La Niña conditions (ENSO cold phase), initiated flash flooding on a country already grappling with disaster. This was exacerbated by the topography of the country and the addition of rainfall from the remnants of tropical cyclones. A total of 74 provinces were affected, and US $536.6 billion lost due to damages caused by the floods [19]. This intense drought behaviour punctuated by flash flooding, is recognised in the country's history. Paleo-megadroughts, which are severe prolonged droughts, occurred in the then Khmer civilisation over 1340 to 1360 and 1400 to 1420 [20], and were interrupted by flash flooding. This extreme climatic variability is believed to be a cause of the collapse of the Khmer civilisation [21]. Again, in 1756-1758, droughts interspersed by devastating flash floods, resulted in agricultural turmoil, great enough to trigger political uprising and reform in the country [22]. It is clear these climatic changes leading to droughts and/or flooding cause high levels of agricultural ruin. Despite the focus in recent years on creating a more modern economy, agriculture is still integral to Thailand's economy. It is a key exporter of widely consumed products such as rubber, sugar and rice. Yet despite the country's susceptibility to climatic variation, the cause and impacts of monsoon variability over Thailand have not been as intensively studied as over the Indian sub-continent or China [1]. Improved understanding of this relationship could have huge benefits in predicting and mitigating against these major hydrological events, leading to positive societal change in the country. However, to predict future behaviour, a complete and long term knowledge is needed of both monsoon variability and the range of influencing climatic controls from the tropical Pacific. This is most efficiently achieved from a paleoclimate perspective [24]. Much of the evidence for these megadroughts across Thailand and SE Asia comes from proxy record collections, including speleothems [25], ocean sediments [26], corals [27] and ice cores [28]. Using a range of proxies, links between hydrological disasters and the climatic controls triggering them can be understood.
This study will use proxies from the last 2400 years to investigate the relationship between the occurrence of ENSO and droughts in Thailand. The paleoclimate, in particular drought conditions and rainfall deficiency, will be determined from speleothem, lake sediment and tree ring records. These proxies were taken from three independent studies [29][30][31] and are sourced from varying locations, as illustrated in Figure 2. The proxies will be plotted against ENSO conditions. ENSO warm events are determined from modelled ENSO data over the last 2400 years and recorded historic observational data from 1870 onwards.

Tree Rings
The original tree ring data for Thailand was collected and presented in a study by Cook et al. [29]. This study collected tree ring growth data for a large region within Asia to create the Monsoon Asia Drought Atlas (MADA) and later analysed it to reconstruct the regional climate history in the form of the Palmer Drought Severity Index (PDSI) [29]. As tree ring data is limited in how far it extends back in time, the authors reconstructed the regional history for the past 700 years. The findings illustrated periods of four significant droughts: the Ming Dynasty Drought (1638-1641), the Strange Parallels Drought (1756-1768), the East India Drought (1792-1796) and the Great Drought (1876-1878) [29]. From these data and models, monsoon behaviour can be better understood and predicted for future generations.
Tree ring radial width and age data was obtained from a study by Cook et al. [29] for five sites in Northern Thailand. This data was collected through field sampling and is part of a larger collaborative effort to produce an extensive tree ring chronology within Asia [29]. In Thailand the species was Pinus merkusii and data pertaining to it was collected from Cook et al. [29] and accessed through the National Oceanic and Atmospheric Administration (NOAA) Paleoclimatology Database. This tree ring data provided yearly and decadal timespan data, which illustrated the effects of drought as it relates to ENSO. Within this tropical region the growing season (ring growth) occurs during the monsoon season, between May and October [32]. Tropical trees can grow year round since the winter season is not significantly different. The Regional Curve Standardisation (RCS) method was used to standardise the yearly average ring width for each site and the region. Standardisation allows for data to be studied without the bias of growth factors. An average of the tree ring chronologies was then plotted against historical ENSO and stalagmite data ( Figure 3). One potential bias that presently occurs within the Pinus merkusii species in Thailand, which is burnt by the hill tribes leaving false rings and unclear ring boundaries thereby impacting the regional ENSO interpretation [ [31]. This stalagmite data was also compared with volcanic records, Northern Hemisphere temperatures and marine sediment proxies. With all of this information, Tan et al. [31] created an ITCZ shift index for this 2700 year period which illustrates how the ITCZ shifts over time [31].
The speleothem proxy is taken from Tan et al. [31]. The record spans 706 BC to 2004 AD, and is determined from three stalagmites from Klang Cave ( Figure 2). The ratio of δ 18 O to δ 16 O in 1388 samples of the stalagmites, infer paleo-conditions as the isotopic ratio is dependent upon changes in rainfall. This is proven by a negative correlation with the proxy to Bangkok rainfall from 1901 to 2004 (r = −0.52, p < 0.01) [31]. The stalagmites studied recorded rainfall variations on decadal to centennial timescales and climatic conditions within Klang Cave allowed for consistent annual stalagmite growth [31]. The data was access through the NOAA Paleoclimatology Database and is plotted against modelled ( Figure 4) and historical ENSO measurements ( Figure 3).

Lake Pa Kho Sediment Core
The study by Chawchai et al. [30] analysed the sediment core data from Lake Pa Kho over the last 2000 years to better interpret climatic shifts and events. The data set used was divided into lithostratigraphic, geochemical (i.e., charcoal and carbon) and biological (i.e., diatoms and plants) sections based on the data studied. These sets of data were then combined to form a reconstructive history of the region as it relates to climate shifts. Chawchai et al. [30] separated the lithostratigraphic unit into different sections based on plant productivity and how this related to the summer monsoon season. The climatic reconstruction was determined from measurements of total organic carbon percentages, C/N ratios, δ 13 C isotope levels, biogenic silica (BSi) totals, charcoal presence, plant composition and silica/potassium/titanium levels. Climatic reconstructions were presented to assist in analysing past hypotheses regarding decadal to centennial variability in monsoon climates. These hypotheses surrounded ENSO, solar forcing influence, ITCZ shift, Pacific Walker Circulation and Indian Ocean Dipole (IOD) [30].
Two overlapping 10 m sediment cores from Lake Pa Kho provide a record of the region's paleoclimate. The study by Chawchai et al. [30] analyses the section spanning 2.00 to 3.50 m in depth, the results of which are used in our research. The sediment record was split in 1 cm intervals and radiocarbon dated to ages 177 BC-2010 AD. However, there is a large hiatus present from 970 AD to 1300 AD. The samples were then geochemically and elementally analysed; our study will focus on the resulting fluctuations in δ 13 C ( Figure 4). The stable carbon isotope of δ 13 C is a proxy indicator for changes in precipitation and humidity based on the influence of aquatic or terrestrial plants [30]. The variables recorded in the study by Chawchai et al. [30] focus primarily on the summer monsoon season and how the ITCZ shifts as a result. Therefore, the δ 13 C isotope recorded represents summer precipitation within the region on decadal to centennial time scales. This lake site is located at the convergence of several climate phenomena making it difficult to attribute all activity to simply ENSO cycles.  [31], the grey lines represents the raw data which is overlaid by black smoothed data using the Savitzky-Golay filter (R 2 = 0.25). (b) Sediment core δ 13 C values [30]. (c) Modelled SST for Niño region 3.4 spanning from the TraCE-21ka simulation (sourced from Earth System Grid: https://www.earthsystemgrid.org/project/trace.html). Blue shaded areas represent drought periods discussed by the papers associated with each proxy. The grey shaded area shows the hiatus in the sediment core. Trendlines are present to show the overall gradual change to a drier climate.

Proxies as a Record of Drought
A PDSI was calculated for the area and exhibited alongside the proxy data; this is presented in the Supplementary Information, Figure S1, to outline how the proxy data used aligns with a drought index. The PDSI illustrates drought severity as a relationship between the amount of precipitation required for the region to sustain itself and the actual amount of precipitation [33]. For the purpose of this study, drought indices show how ENSO alters drought severity and conditions over time. Overall, the PDSI and the proxy data exhibit a drying trend over the tested timespan.
The proxy data was verified with other proxy data, however, none were available for Thailand and so alternate speleothem data was taken from Laos [34], and lake sediment data from southwest China [35]. These were compared with the proxies, the result is shown in Supplementary Information Figure S2. There are similarities between the lake sediment cores, both recognising drought periods across 600 to 900 AD and a sudden climatic shift to drier conditions around 1300 AD. Less similarities are seen between the speleothem proxies in terms of overall trends, but certain climatic events are recognised in both. We take this as verification of the proxies.

Modelled ENSO Reconstruction
Results from the Transient Climate Evolution (TraCE)-21ka climate simulation, conducted with the Community Climate System Model version 3 (CCSM3) [36], were used to compare modelled ENSO variability with proxy data for the last 2 ka. The TraCE simulation uses a set of realistic climatic forcings: orbital variations, greenhouse gases, continental ice sheets and meltwater discharge, and has been shown to replicate many aspects of climatic shifts over the past 21 ka [36,37]. Sea surface temperatures (SST) for the Niño 3.4 region over the central-eastern Pacific Ocean (5 N-5 S, 170 W-120 W) were extracted to represent the average SST anomalies across the central equatorial Pacific [38]. Once extracted, a monthly field mean was calculated for our Pacific region followed by a 50 year running mean over the period 440 BC to 1964 before being detrended ( Figure 4). This smoothing of the SST was done to achieve a better trend in the climate, more appropriate for comparison with other data. It is important to state that this does not explicitly show when an El Niño (or La Niña) period occurs, but indicates more favourable conditions for stronger ENSO events. The index may also capture other anomalous climate patterns, perhaps in the absence of ENSO events for example Pacific Decadal Oscillation (PDO) [39,40].

Historic Measured ENSO
The Multivariate ENSO Index (MEI.v2) identifies in situ measured ENSO anomalies. It is a composite scale, with the original MEI.v2 (1979 to present) based on five variables: sea level pressure, SST, zonal and meridional components of surface wind and outgoing longwave radiation [41]. This is extended back to 1871 using sea level pressure [42] and SST [43]. MEI.v2 index greater than 0.5 (lesser than −0.5) is classified as El Niño (La Niña) conditions. The data was sourced from NOAA's Earth System Research Library. To ensure the modern (1979 to present) MEI index is capturing ENSO and not other components of variability such as PDO, it was correlated against monthly sea surface temperatures for the Niño 3.4 region, determined from the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST) [44]. A strong positive correlation (r = 0.76, p < 0.01) indicates that the MEI.v2 index is showing ENSO variation. A further correlation was performed against MEI with a decadal running mean of Hadley SST to remove any possible low frequency signal of PDO. Another strong positive correlation (r = 0.75, p < 0.01) indicates that PDO only contributes to a small fraction of the correlation and would be more influential over centennial timescale. Two-yearly averages of MEI anomalies were taken to statistically compare ENSO with proxy climate data; a 12 month average starting from May, encompassing the onset of El Niño, and a 3 month average from November to January, capturing the peak [8].

Comparison to Modelled ENSO
Climatic proxy data for both the speleothems and sediment core ( Figure 4) show increasing trends. For the speleothem sequence, δ 18 O values are seen to be rising over the past 2400 years, indicating the region is experiencing a gradual reduction in precipitation. The proxy data attained from the sediment core draws a similar conclusion, with δ 13 C values decreasing throughout the sequence. Significant peaks in both proxies infer possible drier conditions and droughts.
Blue shadings in Figure 4 indicate periods of perceived drought as discussed in Tan et al. [31] study of the speleothem sequence and Chawchai et al. [30] study of the sediment core. It appears that the perceived years of drought differ between the proxies. A sharp rise in δ 13 C found in the sediment core between 370 and 410 AD followed by 450 and 800 AD is not reflected in the speleothem's chronology. In the speleothem a decrease in δ 18 O levels between 950 and 1300 AD, split by a 50 year period of much wetter condition between 1150 and 1200 AD, reflects a period of much drier conditions known as the Medieval warm period. This is mirrored by a large hiatus in the sediment core record spanning from 970 and 1300 AD. Chawchai et al. [30] infers this hiatus to be a result of low levels of accumulation caused by the decomposition of material due to aerobic conditions, a reflection of much drier conditions. Further evidence is given by the low δ 13 C values (28%) immediately following the hiatus, and continuing until 1450 AD suggesting the expansion of terrestrial plants and decreased moisture. If true, the hiatus would mirror to the drier period found in the speleothem at the same time. However the period of drier conditions found in the sediment core after the hiatus is not represented in the speleothem.
The SST data presented in Figure 4 is highlighted red for anomalies above 0 • C and blue for below 0 • C, representing the possibility for El Niño or La Niña like conditions, respectively. Based on this, the model shows El Niño like conditions dominating 300 to 100 BC, at the same time the speleothem sequences shows no noticeable drier periods and instead decreasing δ 18 O values. The models follows this phase with much cooler SST leading up to 0 AD and possible La Niña-like conditions, this corresponds to increasing δ 18 O values in the speleothem and a drier period. The first noticeable period of inferred drought conditions found in the sediment core at 370 to 410 AD mirrors El Niño-like conditions found in the simulation. However, the large dry period from 450 to 800 AD inferred from the sediment corresponds with La Niña dominated conditions. Subsequently, the model shows a period of more dominated warm SST and possible El Niño condition from~1000 to~1400 AD, this corresponds with drier periods inferred from both the speleothem and sediment core.

Comparison to Historical Measured ENSO
Comparing the speleothem sequence and tree ring proxies against the much shorter time scale of MEI.v2 ENSO index from 1871 to 2006 (Figure 3) reveals similar trends to those seen in Figure 4. Standardised tree ring width is shown to be shrinking from~1.2 to~1.0 between the beginning of time series to the 1980s, indicating harsher growing conditions and an inferred reduction in precipitation. However, this is followed by a sharp increase of values to~1.6 in 2005. The speleothem sequences are similar; δ 18 O undergoes an increase from~−5.5 to peak around 1930s, inferring decreased precipitation, and then continues on a small downward trend following this. The two proxies show a moderate but significant correlation of r = −0.373 (Table 1). The tree ring proxy does not undergo the downward trend as seen post 1930 in the speleothem record, thus the negative correlation value.
Historic ENSO variations spanning from 1871 to 2006 illustrate the cyclical nature of this climatic event ( Figure 3). Moving through the ENSO sequence, a large La Niña event is shown at 1910-1912, this corresponds with an inferred increase in precipitation from the speleothem, but it is contrasted by tree ring width which is shown to decrease. The lowest δ 18 O values in the speleothem, implying the driest conditions, occur alongside SST anomalies surrounding 1930, an event not identified by Seager et al. [39] but still a recongised El Niño. This is followed by a marked increase in δ 18 O levels, implying increasing wetter conditions, which correspond to a lack of El Niño activity leading up to 1939. The 1939-1940 El Niño does not appear to visibly impact either proxy. Where there is clustered or strong ENSO activity occurring on Figure 3, this may indicate PDO influence [39][40][41]; however, this is considered a background affect [45], and it is still possible to identify the El Niño and La Niña events. The 1982-1983 El Niño is the most volatile depicted on the MEI index. It corresponds to the lowest standardised tree ring width signalling a stressful growing season but no noticeable changes are seen in the speleothem record.
A visual interpretation of the historic ENSO events and data obtained from the climate proxies leaves any relationship hard to determine. The results of a Pearson's correlation test between 3 and 12 month averages and historical proxy data (Table 1) show speleothem and tree ring data have a minimal and insignificant correlation to ENSO averages.

ENSO Climate Impacts
How El Niño impacts east Asian climate, such as that of Thailand, needs to be reflected on when understanding ENSO's contribution to drought or climate disruption. While the peak of ENSO and the materialisation of its impacts occur during Boreal winter [46], anomalous ENSO behaviour is apparent after peak ENSO phase, and contributes to altering usual climate. As highlighted by Wang and Zhang [47], the anomalous Philippine Sea anticyclone (PSAC) is one integral system transferring El Niño influence onto the East Asian climate. Prior to an El Niño phase, during the boreal summer, El Niño stimulated central Pacific convection anomalies are responsible for the generation of cyclonic circulation anomalies in the Philippine Sea. This increases wind speeds and drives colder SST anomalies, providing suitable conditions for the formation of Philippine high pressure anomalies. This is responsible for weakening the east Asian winter monsoon during the ENSO peak, but then bringing abundant rainfall in the following spring to summer to the east Asian monsoon front. With further examining ENSO, the anomalous behaviour that accompanies it must be considered. While the peak of El Niño, and the detrimental effects, occurs during the boreal winter and then sharply declines in spring, anomalous atmospheric circulation may be responsible for extending the influences of ENSO to impact summer monsoon behaviour. Work by Ronghui et al. [48] and Xie et al. [49], show ENSO related anomalous remain during the following summer, with Northwest Pacific SST anomalies much greater than that prior to ENSO whilst Eastern Pacific SST anomalies have become much weaker. These persisting ENSO summer anomalies may be due to regional anomalous anticyclonic circulation [49] mainly the El Niño induced Tropical Indian Ocean warming, which sustains atmospheric anomalies even after the dispersal of El Niño and suppresses rainfall over the subtropical northwest Pacific. These persisting post-El Nino summer anomalies may play an important role in how climatic impacts such as droughts are understood, and therefore should be taken into consideration when interpreting the climate record.

Drought Trends within the Data
The results presented in Figures 3 and 4 indicate ENSO anomalies are not the sole driver of droughts in Thailand. Using historic evidence of known droughts, it seems the proxies do not accurately represent climatic changes. It is widely established that a prolonged drought over 1340-1360 and 1400-1420 violently affected the Khmer empire [21]. Drought conditions at the turn of the 13th century are visible from the lake sediment δ 13 C record in Figure 4, but the speleothem record shows no diversion from the norm. During the widespread megadrought in SE Asia over 1756-1758 [22], there is no fluctuation on Figure 4 in neither the speleothem nor lake sediment proxies. In Figure 3, the El Niño that likely triggered the megadrought of 1876-1878 [29] is visible, but again there is no evidence of drought conditions that were prevalent across SE Asia.
Looking at long-term climatic variance, moving from the decadal variation associated with ENSO, to millennial, it is hard to establish any trends other than overall decreasing precipitation. Climatic periods, such as the Little Ice Age (LIA),~1300s to~1800s [31], have been clearly established in published work concerning proxy data. Speleothem records from Danda Cave, India, identify recurrence of drought throughout the LIA [25] and Cobb et al. [20] showed persistent drought conditions during the LIA, spread across the tropical Pacific. Figure 4 however, does not show any climatic variance through the LIA.

Errors Associated with Proxies and Inaccuracies of ENSO Observations
The lack of climatic reflection in the proxies could be encouraged by errors associated with the proxies, with the errors surrounding the proxies from the original studies transferred to this work. Speleothem as a proxy has been shown to have differing levels of accuracy when recording ENSO conditions [50]. This is likely due to the wide range of variables affecting δ 18 O, such as cave and air temperature, distance travelled, evaporation [50] and changes in seasonal rainfall [51]. The speleothem data obtained from Tan et al. [31], has dating uncertainties of ±2 to 4 years for stalagmite TK16 (growing from 2 AD to 998 AD), <10 years for stalagmite TK133 growing between 706 BC and 1867 AD, and as low as ±0.4 to 2 years for the TK131 stalagmite growing from 1733 AD to 2004. Using δ 13 C from lake sediment data to determine the timing of climatic events is still a relatively new proxy. ENSO teleconnections have previously been established [52], but the associated errors, such as mixing of laminae, bioturbation, runoff and multiple deposition reduce the reliability of the proxy [53]. The lake sediment data adopted from Chawchai et al. [30] had relatively small 14 C date error margins. However, dates for 2.66 to 2.63 m and 2.63 to 2.60 m differ in age by c 460 calibrated 14 C years which was explained by the presence of a hiatus, as used in this study, but could also be a product of low production rates, increasing the uncertainty of the lake sediment timeline. Tree rings, while shown a successful Holocene drought indicator by the authors of Cobb et al. [20], have historically been a tricky proxy to use in tropical settings [54]. The sample adopted in this study was fraught with issues, including difficult interpretation of the wood of the species used, P. merkusii, and missing samples due to deforestation by indigenous communities [32]. A new growth of trees, in the 1980s, also may have skewed overall tree ring width, despite standardisation.
Alongside problems with the proxies, using modelled or historic ENSO data may also not be sufficient to draw conclusions. Modelled data only simulates SST whilst the historic MEI.v2 index only indicates oceanic and climate anomalies [13], neither show actual events. No two El Niños are the same, and thus the same variables cannot be used to pinpoint every event, nor recognise its strength. Even when historic evidence is present, it is still difficult to define if conditions are actually ENSO. The El Niño of 1918 is a key example. Here, distinct El Niño teleconnections were recognised across the globe: weak hurricane systems, failure of the Indian monsoon and droughts in Australia [55]. Yet SST patterns in the tropical Pacific were regarded as tepid [56], meaning until recent modelling work examining ocean atmosphere interference [57] the El Niño was regarded as weak. On Figure 3 while El Niño like conditions are evident for 1980, it is not an abnormally strong event, but does respond to more drought favouring conditions in the proxies. Another large El Niño event of 1997/98, argued to be the strongest climate anomaly of the 20th century with huge associated anomalous monsoon behaviour [58], is also evident on Figure 3. Yet this does not correspond with dry conditions within the proxies despite El Niño related droughts recorded across Southeast Asia during this period [59]. As discussed previously, no El Nino events occurred between 1931 and 1941, yet the index on Figure 3 suggests otherwise. This is an indication of the index's ability to also capture other events, such as PDO [39,40]. For the CCSM models there could be the possibility of a cold tongue bias. A cold tongue bias can occur in CGCM models and is associated with the double-intertropical convergence zone (ITCZ) problem, where there is excessive precipitation off the equator and insufficient precipitation on the equator in comparison with observational data [60]. As shown by Li et al. [61], CGSM models have one of the lower cold tongue biases than that of CMIP5 models. For CCSM3 a 1990 run by Yu et al. [62] showed that the biennial monsoon variability is very effective in exciting biennial ENSO variability in CCSM3. This is via the mechanism where the monsoon variability excites an anomalous surface wind pattern in the western Pacific. This projects well into the wind pattern associated with the onset phase of the simulated ENSO, but due to the biennial frequency there may be slight underestimation by the CCSM3. The Yu et al. [62] paper also showed the warm SST bias in the tropical Indian Ocean also increases ENSO variability, through encouraging greater mean surface easterlies along the equatorial Pacific, in turn strengthening the Pacific ocean-atmosphere coupling and promoting ENSO intensity.

Other Causation of Droughts
Previous work examining the relationship between smaller scale or localised droughts and ENSO, have drawn similar inconclusive results. Work by Buckley et al. [24], centred on decadal scale droughts in Thailand identified from tree ring chronologies, deemed that the cause of droughts was uncertain, with ENSO being not the sole factor influencing climate. Work by Singhrattna et al. [1], which examines if Thailand monsoon variability is a function of ENSO, found that a correlation between the two was only significant post 1980. Pre-1980, the study found ENSO was instead correlated with the Indian Ocean monsoon variability. While El Niño could be a factor in causing Thailand droughts, it is not the sole component. Other factors are likely to be driving monsoon variability alongside ENSO to create drought episodes. Land surface processes, such as snow cover and vegetation structure across Northern Asia, may be influencing anomalous monsoon behaviour, via modulating land-ocean gradients and ENSO teleconnections [63]. Kumar et al. [18] shows evidence that the Walker circulation is the dominant mechanism to influence rainfall fluctuation in Thailand. This is based on the theory that ENSO presence constrains the descending branch of the Walker circulation within the tropical Pacific, causing less convection in the tropical western Pacific regions and affecting rainfall in Thailand. ENSO's control on SE Asian climate may very likely be modified by other climate drivers, in particular the IOD, which is likely coupled with ENSO over the last millennium [64]. Recent work by Abram et al. [64] illustrates the connection between the two interlinked phenomena and drought exhibited by the huge climate variance of the 17th century caused by strong El Niño variability, weak Asian monsoon and anomalously strong IOD variability. McGregor and Ebi [65] provides further evidence of this, showing the amplifying effect of El Niño when it is in phase with the IOD, and Bird et al. [66] reiterates this, exhibiting the role of IOD variability in driving Holocene Indian summer monsoon variation.
Further to these additional climate patterns, the drought ENSO teleconnection may also be affected by natural volcanic forcing. Analysis by Brad Adams et al. [67] shows volcanic eruptions may provide a subtle forcing of the ocean atmosphere system to a state favouring El Niño conditions, with a possible doubling of the likelihood of an El Niño occurring in the winter following a tropical volcanic eruption. To understand if this was a possible causation to droughts in Thailand, we compared our modelled data to aerosol optical depth values for the past millennium for tropical (0 to 30 • N) volcanic eruptions [68], evident in Supplementary Information, Figures S3 and S4. There is mild evidence of driving force of volcanic forcing coinciding with drought period. Eruptive activity over 1200 to 1300 AD aligns with more El Niño conducive conditions and corresponding dry periods, and so do select eruptions over the 20th century such as over 1900 to 1905 and 1990 to 1995. This indicates volcanic forcing may be a disruptor of ENSO teleconnections for some drought periods, however no standout causal relationship emerged indicating the outcome of volcanic forcing on the proxy data used is unclear.

ENSO in a Changing Climate
It is unclear how climate patterns such as ENSO will change in a warming world. It is a complicated give and take process between different atmospheric components, each affected individually by climate change. The timing of climatic patterns may be shifted, such as El Niños occurring earlier in boreal winter time [34]. ENSO events could increase in frequency [61,69] and/or intensity [70]. A change in the Southern Oscillation means the ascending branch of the Walker Circulation moves westward towards Thailand, enhancing monsoonal rains [17]. Fundamentally, global warming is likely to lead to a more unpredictable climate which will have detrimental repercussions on Thailand's agriculture, both through the climate anomalies accompanying a change in ENSO and the general impacts of global warming: rising temperatures and coastal inundation. Extended droughts or flash floods will exacerbate problems the country is already experiencing, due to environmental neglect, poor infrastructure and Bangkok-centric development policies [71]. Tensions already present surrounding water resources, stemming from dam projects along the Mekong river reducing the water level [72], will increase. The agricultural sector, and the millions that depend on it, are particularly vulnerable to these unpredictable changes. Rice farmers, especially poorer communities, in Thailand were shown by Felkner et al. [73] to be unable to neutralise the adverse effects of extreme climate change, such as sudden drought conditions forced by El Niño. Agricultural and economic losses in the farming community can be reduced by implementing even simple mitigation strategies. Following the 2015 El Niño, which decimated southeast Asian growing seasons, initiatives were launched in Papua New Guinea to distribute seedlings for hybrid drought resistant crops that can better cope with drought or frost conditions [74]. Taro/sweet potato/yam is more likely survive harsher than normal drought conditions, whilst vegetables such as broccoli, cauliflower and cabbage can sustain colder temperatures from frost. This initiative is applicable to other nations like Thailand affected by El Niño, with these El Niño resistant plants first directed to the most at risk areas.
If the majority of agriculture however remains focused on rice production, there are strategies employed across India that have enabled rice production to remain despite fluctuations in the monsoon. Rice crops aided with mid season drainage technology [75] and using alternate wetting and drying [76] are two methods that can sustain rice production. Under a changing climate, hydrological disasters are likely to impact further than just the agricultural sector, encroaching on industrial sectors, tourism and human consumption supply. To cope with this, there needs to be a large adjustment in water consumption behaviours, a reduction in usage, and policy implementation at a government level. Social safety nets need to be in place to cope with incoming climatic disasters, including workfare programs, cash transfers and provision of disaster insurance [77]. The stockpiling of water by local governments, which creates free access for farmers if drought conditions prevail, needs to be properly invested. Overall, the country needs to shift to a water demand strategy, rather than the current water supply strategy, to try to eliminate water loss.

Conclusions
El Niño is likely not the sole driver of drought conditions in Thailand. The confluence of numerous climatic patterns are instead likely to cause the irregular climate behaviour responsible for large scale hydrological events. To further narrow down how climate patterns are influencing droughts, more proxies need to be studied. The inclusion of organic matter-diatoms, pollen and coral-may give better representation of paleo-drought conditions, as would the use of multiple sites. However this depends on the availability of accurate proxies, of which Thailand is lacking. Therefore expansion to include sites in Vietnam, Laos and Cambodia may be needed. It is imperative future research on this subject occurs. The unpredictability of climate behaviours under a changing climate is likely to significantly impact millions of people through hydrological events. Furthermore, population rise and increased demand for land and water resources will put immense pressure on the already tested water supply in Thailand. Mitigation policies are essential moving forward into an unknown climate.
Supplementary Materials: The following are available at http://www.mdpi.com/2571-550X/3/2/18/s1, Figure S1: Comparison between the PDSI and the Speleothem and Lake Sediment proxy record; Figure S2: Comparison of Speleothem proxy records from South East Asia; Figure S3: A plot comparing climate proxies and modelled SST anomalies from 425 BC to 1964 AD; Figure S4: A plot comparing historic ENSO events to climatic proxies.

Acknowledgments:
The model data analysis were performed by resources provided by the Swedish National Infrastructure for Computing (SNIC) at the National Supercomputer Centre (NSC) partially funded by the Swedish Research Council through grant agreement no. 2016-07213.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript: