Multi-scale drought and ocean–atmosphere variability in monsoon Asia

Spatially extensive and persistent drought episodes have repeatedly influenced human history, including the ‘Strange Parallels’ drought event in monsoon Asia during the mid-18th century. Here we explore the dynamics of sustained monsoon failure using observed and tree-ring reconstructed drought patterns and a 1300-year pre-industrial community earth system model control run. Both modern observational and climate model drought patterns during years with extremely weakened South Asian monsoon resemble those reconstructed for the Strange Parallels drought. Model analysis reveals that this pattern arises during boreal spring over Southeast Asia, with decreased precipitation and moisture flux, while related summertime climate anomalies are confined to the Indian subcontinent. Years with simulated South Asian drying exhibit canonical El Niño conditions over the Pacific and associated shifts in the Walker circulation. In contrast, multi-year drought periods, resembling those sustained during the Strange Parallels drought, feature anomalous Pacific warming around the dateline, typical of El Niño Modoki events.


Introduction
The Asian monsoon system is a large-scale coupled ocean-atmosphere phenomenon characterized by distinct seasonal precipitation and circulation features. Variability in monsoon rainfall has a significant influence on economic and agriculture production across tropical Asia. Remote oceanic influences are known to modulate the monsoon system, including the El Niño Southern Oscillation (ENSO) and Indian Ocean dipole (IOD; Ashok et al 2004, Ummenhofer et al 2011and references therein), resulting in extensive and sometimes persistent droughts (Wang 2005. Pacific sea surface temperature (SST) anomalies associated with El Niño events affect the Asian monsoon through an eastward shift in the zonal Walker circulation, and anomalous subsidence over monsoon Asia (Rasmusson andCarpenter 1983, Ummenhofer et al 2013). More recently, variations in the location of maximum warming in the equatorial Pacific during El Niño events have been associated with regionally distinct effects on Asian monsoon variability. Canonical (eastern Pacific) El Niño and so-called El Niño Modoki events have different spatial and temporal signals , Taschetto et al 2010. El Niño Modoki events, characterized by a warming pattern in the central Pacific, influence tropical circulation via meridional SST gradients and an anomalous two-cell Walker Circulation in the troposphere throughout the tropical Pacific , Taschetto et al 2010, Tao et al 2014.
Instrumental and paleoclimate observations reveal anomalous monsoon variability across a range of spatiotemporal scales, including both interannual-and decadal-scale drought and pluvial periods (e.g. Yihui and Chan 2005, Wang et al 2008. Severe and persistent droughts in the past in monsoon Asia are evident in tree-ring paleoclimate reconstructions, particularly from the spatiotemporal Monsoon Asia Drought Atlas (MADA). The MADA utilizes a network of more than 300 annuallyresolved tree-ring width time series covering the last 700 years . 'Megadroughts' of greater intensity and duration than those in the instrumental record are evident in these paleoclimate reconstructions, including the particularly severe, extensive, and persistent 'Strange Parallels' period (1756-1768 CE, figure 1(a)), a monsoon failure that coincided with societal upheaval and the collapse of polities in Thailand, Myanmar, and Vietnam . While skillful representation of Asian monsoon variability remains a challenge for state-of-the-art general circulation models (GCMs; Turner and Annamalai 2012), paleoclimate field reconstructions, like the MADA, provide out-of-sample evaluation targets for climate models, particularly of their ability to simulate decadal-scale hydroclimate features that have societal relevance. For their part, GCMs further our understanding of the coupled ocean-atmosphere dynamics that bring about prolonged drought episodes.
We can compare the spatially broad and temporally persistent Strange Parallels drought with observed Palmer Drought Severity Index (PDSI) values (1901Dai 2011) during periods of weak monsoon circulation. Years of weakened monsoon circulation are identified using the South Asian monsoon index (SAMI), a broad-scale circulation index encompassing regions over India and Southeast Asia (70°E-110°, 10°N-30°) that measures the meridional wind shear between the lower and upper troposphere (V 850 -V 200 ) (Goswami et al 1999). Composites for both paleoclimate and observations are the mean PDSI values over the boreal summer monsoon season (June-August, JJA). A MADA composite of years with weakened monsoon during the instrumental period ( figure 1(b)) reveals a similar spatial drought pattern, as that reconstructed for the Strange Parallels drought (figures 1(a), and 2(b) in Cook et al 2010): i.e., with dry conditions in the Indian subcontinent and Southeast Asia, and wet anomalies over northern portions of the Tibetan plateau and northwest China. The instrumental PDSI also broadly reflects this drought pattern (figure 1(c)). The magnitudes seen in the MADA and instrumental PDSI differ as tree-ring reconstructions may only capture a fraction of the instrumental variance and the combination of reconstruction ensembles can further reduce the amplitude of the signal .
In this paper, we assess how weakened large-scale South Asian monsoon circulation leads to both interannual and decadal-scale drought conditions, and how these are linked to the ocean-atmosphere dynamics of the Indo-Pacific system. As seen in figure 1, weak monsoon years during the instrumental period exhibit drought patterns across monsoon Asia that are reminiscent to that found during the Strange Parallels drought in the MADA. We examine the largescale climatic conditions that could have given rise to the Strange Parallels drought, and what this reveals about the dynamics of ocean-land-atmosphere interactions as they relate to the Asian monsoon system. Given the decadal and multi-decadal nature of the Strange Parallels drought in the 18th century, the instrumental record lacks the temporal coverage necessary to explore low-frequency variations of the South Asian monsoon on such decadal to multi-decadal timescales. Instead, we use a long simulation of the National Center for Atmospheric Research (NCAR) Community Earth System Model (CESM) to identify and analyze monsoon drought patterns at both interannual and decadal time scales.

Datasets and methodology
The MADA (Cook et al 2010) is a tree-ring reconstruction of the boreal summer (JJA) PDSI, a common index of meteorological drought incorporating both precipitation and evapotranspiration used to quantify long-term changes in aridity over global land areas (Dai 2011). The MADA covers the land area across the domain 60°-145°E and 10°S-58°N at a spatial resolution of 2.5°latitude/longitude and spans from 1300 to 2005 CE. In addition to the tree-ring reconstructed PDSI, we also investigate the observed PDSI during the instrumental record (Dai et al 2004).
A series of monthly gridded observational and reanalysis products are used to assess the large-scale circulation of the South Asian monsoon. These include: National Center for Environmental Prediction (NCEP)/NCAR reanalysis (NNR, Kalnay et al 1996Kalnay et al , 1948Kalnay et al -2011, the 20th Century Reanalysis (Compo et al 2011(Compo et al , 1871(Compo et al -2008, and the European Centre for Medium-Range Weather Forecasts ERA-40 Reanalysis (Uppala et al 2005(Uppala et al , 1958(Uppala et al -2011. We use the SAMI as a measure of monsoon circulation strength, which has previously been employed to assess monsoon strength and evaluate GCM performance (Goswami et al 1999).
We utilized a pre-industrial control run of 1300 years length of the NCAR CESM to understand the dynamics of South Asian monsoon variability across a range of timescales. The CESM is a state-of-the-art GCM that consists of atmosphere, land, ocean, and sea ice components (Gent et al 2011, Meehl et al 2013. The Parallel Ocean Program version 2 (POP2) has 60 vertical levels with a 1.11°zonal resolution and a meridional resolution of 0.27°at the equator, increasing to 0.54°around 33°N/S, and becoming constant at higher latitudes (Gent et al 2011). The Community Atmospheric Model version 4 is the atmospheric component of the CESM with a uniform spatial resolution of 1.25°× 0.9°(Gent et al 2011). We assess the monthly climatology from the CESM against different reanalysis products using the SAMI to ascertain the model's skill in representing South Asian monsoon circulation characteristics, before applying it to analysis of monsoon drought. In addition to model fields simulated in CESM, we calculate a model synthetic PDSI (Dai et al 2004) using simulated precipitation and temperature for direct comparison with the MADA.
We conducted a time series analysis of the simulated JJA SAMI from the pre-industrial control run of the CESM in order to identify years of persistent weakened circulation. We classified years falling below the lowest 10% SAMI threshold (blue dotted line) as years with a weakened monsoon index (n = 130). We also utilize an 11-year running average time series (superimposed onto the raw time series) to identify the decadal signal and capture persistent weakening in the monsoon as decadal local minima (n = 13; lowest 10% of average decadal SAMI values). We then calculate composites of model-derived JJA PDSI in order to assess the model's ability to simulate decadal-scale drought patterns. We evaluated the proximate and large-scale climate conditions associated with simulated monsoon drought by creating composite anomalies for these epochs of weakened monsoon circulation. We created model composite anomaly maps of total precipitation, 850 hPa-level winds, SST, and velocity potential at 200 hPa level for periods with weakened JJA monsoon circulation on both an interannual (n = 130) and decadal (n = 13) basis. In addition to examining the core JJA monsoon season, we also evaluated potential preconditioning of subsequent summer soil moisture and drought conditions during the early/pre-monsoon March-May (MAM) season. The land surface conditions prior to monsoon onset are significant in setting the land-sea thermal contrast and driving monsoon circulation.
A two-tailed t test is used to estimate the significance of climate anomalies during periods with weakened South Asian monsoon strength.

South Asian monsoon in CESM
Overall the model simulation of the SAMI seasonal cycle aligns closely with the NNR and ERA-40 during MAM and JJA ( figure 2(a)). The CESM SAMI is weaker than the 20th century reanalysis; however, the latter is anomalously strong compared to the other reanalysis products and the ±1 standard deviation of the CESM SAMI lies within the range of South Asian monsoon variability based on reanalysis products. The SAMI simulated by CESM during October-December is stronger than the three reanalysis products by 1-3 m s −1 , but captures well the monsoon strength during MAM and JJA, our seasons of interest here.
The pre-industrial control run of the CESM simulates periods of weakened monsoon circulation at both interannual and decadal timescale ( figure 2(b)). The interannual CESM PDSI composite mimics the observed spatial drought patterns seen in the paleoclimate and instrumental observational record (figure 1), especially the signal in western India and parts of Pakistan and the excess moisture pattern over the northern regions of the Tibetan plateau ( figure 2(c)). At the decadal time scale, the signal amplitude is reduced as expected, but the same coherent spatial patterns remain, including the horseshoeshaped signal across India and Southeast Asia (figure 2(d)).

Impact of weakened South Asian monsoon on regional climate
We examined both interannual and decadal timescales in order to identify the dynamical changes leading to the associated drought conditions over India and Southeast Asia (figure 3). For the individual years (10%, n = 130) with the lowest simulated SAMI index, figure 3(a) shows an associated anomalous anticyclonic rotation flow over India, with an anomalous southwest flow in northern India and northeast flow in Southeast Asia, indicative of the weakened monsoon during MAM. Modeled anomalous drying of ∼40 mm/month is centered in the southern regions of Southeast Asia including Thailand, Cambodia, Vietnam, and portions of southern India, associated with weakened wind features over the Arabian Sea and South China Sea. On decadal time scales (figure 3(b)), simulated MAM wind anomalies show a similar pattern over Southeast Asia and south India, and anomalous drying of ∼10 mm/month occurs over Myanmar. During JJA for simulated individual weak monsoon years (figure 3(c)), the seasonal crossequatorial flow normally persistent during the monsoonal period weakens considerably. This closely aligns with the simulated anomalous drying pattern of ∼100 mm/month constrained to India, southern Pakistan, and Nepal. In contrast, stronger winds over the Bay of Bengal result in anomalously wet (∼80 mm/ month) conditions over western portions of Southeast Asia. At decadal time scales (figure 3(d)), strengthened and dry northerly winds over northern India coincide with an anomalous drying signal (∼20 mm/month), as well as the above-normal precipitation in Southeast Asia (∼15 mm/month).
SST anomalies and velocity potential at 200 hPa are shown in figures 4(a)-(d). Interannual MAM velocity potential mirrors the anomalous precipitation distributions and SST anomalies; a warm eastern equatorial Pacific drives enhanced upward motion locally, while anomalous subsidence over Southeast Asia favors drought conditions in that region. The MAM SST anomalies are similar to an El Niño signal with an anomalous warm equatorial Pacific (∼0.5°C) and cooling patterns over the Indo-China peninsula. Furthermore, a positive IOD pattern is observed, characterized by warm (cold) SST features in the western (eastern) Indian Ocean, which can be linked to ENSO events through an extension of the Walker Circulation to the west.
Previous studies have shown the CESM to have a reasonable representation of canonical (cold-tongue) and warm-pool (Modoki) ENSO events with regard to the seasonal evolution, magnitude, and patterns compared to observations in a multi-model intercomparison of CMIP5 models (Kug et al 2012, Taschetto et al 2014. A realistic representation of the seasonal evolution of SST anomalies associated with ENSO is crucial for skillfully representing teleconnection patterns across the broader Indo-Pacific region, including monsoon Asia. Furthermore, the CESM also skillfully represents teleconnection patterns to the Austral-Asian monsoon (Jourdain et al 2013).
The decadal signal differs from the interannual, particularly with respect to SSTs. Regions of maximum warming are observed over the central equatorial Pacific with cooler temperatures off the equator in the eastern Pacific, similar to patterns found during El Niño Modoki events where maximum warming is centered over the dateline (e.g., Ashok et al 2007, 2009, Ratnam et al 2010, Taschetto et al 2014. Decade-scale weak SAMI epochs are characterized by anomalous ascent, depicted by the velocity potential, as seen over the region of maximum SST warming and anomalous drying patterns driven by subsidence over Southeast Asia ( figure 4(d)). During individual weak SAMI years ( figure 4(c)), a similar El Niño signal is observed with maximum warming constrained to the eastern Pacific , but with a stronger downward motion over India and Indonesia.
Canonical El Niño events are associated with drought over Southeast Asia, Indonesia, northeastern China and locally over India , which is also seen here in the model composites for individual years with weakened monsoon circulation. During a canonical El Niño event (figures 4(a) and (c)), a shift in the Walker Circulation, along with subsidence over Southeast Asia and anomalous upward motion over the eastern equatorial Pacific, favor a weakened monsoonal flow and reduction in moisture fluxes resulting in drought conditions (Ratnam et al 2010. Interestingly, during individual weak SAMI years ( figure 3(a)), below-normal precipitation is spread throughout south India and Southeast Asia, while below-normal JJA precipitation is confined to India only. This appears to also be consistent with the sensitivity of Southeast Asian tree-ring chronologies to early monsoon season moisture . Previous studies have shown a seasonal phase-locking signal for El Niño including an intensification in late summer to early fall, peak in winter, and decay in the following spring , Tao et al 2014. In contrast, El Niño Modoki events show a different temporal signal, along with remote ocean SST distribution and global impacts. Tao et al (2014) described the differences between the two types of El Niño, including the change in the position of anomalous vertical motion over the Pacific Ocean, which stem from the anomalous Walker Circulation during both types of events, as well as zonal SST gradients in the eastern equatorial Pacific. Furthermore, a less coherent phaselocking signal and large decadal variability is detected for El Niño Modoki . Marathe et al (2015) have shown El Niño Modoki events to differ from the canonical El Niños in their seasonal evolution and teleconnections, including their dynamic and thermodynamic air-sea coupling strength.
Our decadal weak monsoon composites from CESM are associated with an El Niño Modoki with maximum warming in central Pacific SSTs and rising motion over these warmer SSTs and sinking motion over the Indian region that are associated with belownormal precipitation. At annual time scales, anomalous SSTs in the Indian Ocean are much more pronounced due to internal oceanic processes including El Niño-induced Rossby wave propagations (Tao et al 2014). For decadal SST composites, anomalies over the Indian Ocean are less pronounced. This could be a result of a lack of significant wave propagation, like that observed during a canonical El Niño event, causing subtle SST anomalies over the Indian Ocean after an El Niño Modoki event (Tao et al 2014). Other studies have proposed possible mechanisms connecting decadal changes in the ENSO-monsoon system to other components of large-scale ocean-atmosphere variability, including an important role for unforced variability (Li and Ting 2015). The occurrence and persistence of El Niño Modokis in the early 20th century have demonstrated a role of multi-decadal natural climate variability can have in generating El Niño Modokis (Pascolini-Campbell et al 2015). Considerable multi-decadal variability in the frequency of El Niño Modoki events in the observational record has also been described, with El Niño Modoki events more common in recent decades (e.g., Ashok et al 2007, Yeh et al 2011. Such decadal modulation of ENSO characteristics (intensity, spatial distribution, and frequency of the two types) has been attributed to decadal variations in the climate mean state in the Pacific thermocline structure (e.g., Choi et al 2012). However, improved quantification of the effects of decadal variability in the past is needed to determine its future behavior in light of anthropogenic climate change (Turner and Annamalai 2012).

Conclusion
We have identified large-scale climate conditions that can lead to extensive decadal-scale monsoon drought conditions, like those that have been reconstructed using tree-rings during the mid-18th century Strange Parallels drought, an extended period of apparently weak monsoons that may have played an important role in shaping historic regional sociopolitical change (Lieberman 2003. The use of the long pre-industrial control run of the CESM, combined with analysis of paleoclimate drought reconstructions, allows us to develop dynamically-based inferences about causal mechanisms across multiple time scales. Simulated patterns of PDSI from CESM are analogous to those reconstructed for the Strange Parallels epoch and the model critically provides the physical basis for interpreting these rare past events in the context of the large-scale ocean-atmosphere circulation. Our study reveals the influence of two different types of El Niño (canonical El Niño and El Niño Modoki) on drought conditions over monsoon Asia. Canonical eastern Pacific El Niño events can have a substantial influence on interannual variability in monsoon rains; however, decadal-scale drought, like those reconstructed during the mid-18th century, appear to be associated with SST and circulation anomalies similar to El Niño Modoki, which exhibits a larger decadal signal . The different Walker circulation anomalies of the two El Niños result in the distinct patterns in space and time of below-normal precipitation and drought conditions across India and Southeast Asia. Thus, our study provides the foundation for explaining mechanisms that can lead to sustained decadal drought patterns throughout monsoon Asia and confirms the importance of the ENSO characteristics in determining spatiotemporal anomalies of the Asian monsoon system.