This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 129.236.12.90

This content was downloaded on 19/07/2016 at 16:31

Please note that terms and conditions apply.

The improbable but unexceptional occurrence of megadrought clustering in the American

West during the Medieval Climate Anomaly

View the table of contents for this issue, or go to the journal homepage for more

2016 Environ. Res. Lett. 11 074025

(http://iopscience.iop.org/1748-9326/11/7/074025)

Home Search Collections Journals About Contact us My IOPscience

Environ. Res. Lett. 11 (2016) 074025 doi:10.1088/1748-9326/11/7/074025

LETTER

The improbable but unexceptional occurrence of megadroughtclustering in the AmericanWest during theMedieval ClimateAnomaly

SloanCoats1, Jason E Smerdon2, Kristopher BKarnauskas3 andRichard Seager2

1 Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO,USA2 Lamont-Doherty EarthObservatory, ColumbiaUniversity, Palisades, NY,USA3 Cooperative Institute for Research in Environmental Sciences andDepartment of Atmospheric andOceanic Sciences, University of

Colorado, Boulder, CO,USA

E-mail: sloan.coats@colorado.edu

Keywords: climate, paleoclimate,megadrought, North America

AbstractThefivemost severe and persistent droughts in the AmericanWest (AW) during theCommonEraoccurred during a 450 year period known as theMedieval Climate Anomaly (MCA—850–1299C.E.).Hereinwe use timeseriesmodeling to estimate the probability of such a period of hydroclimate changeoccurring. Clustering of severe and persistent drought during anMCA-length period occurs inapproximately 10%of surrogate timeseries that were constructed to have the same characteristics as atree-ring derived estimate of AWhydroclimate variability between 850 and 2005C.E. Periods ofhydroclimate change like theMCA are thus expected to occur in the AW, although not frequently,with a recurrence interval of approximately 11 000 years. Importantly, a shift inmean hydroclimateconditions during theMCA is found to be necessary for drought to reach the severity and persistenceof the actualMCAmegadroughts. This result has consequences for our understanding of theatmosphere-ocean dynamics underlying theMCA and a persistently warmAtlanticMultidecadalOscillation is suggested to have played an important role in causingmegadrought clustering duringthis period.

Introduction

Proxy records suggest that the American West (AW)(125°W–105°W, 25°N–42.5°N; hereinafter AW) hasexperienced multidecadal drought, the severity andpersistence of which is beyond the range of hydro-climate variability observed over the instrumentalinterval (1870-present). While these ‘megadroughts’have been a prominent, albeit infrequent, feature of theCommon Era (C.E.—the last 2015 years) paleoclimaterecord in the AW (e.g. Stine 1994, Stahle et al 2000,2007, Cook et al 2004, 2007, 2010,Herweijer et al 2007),the occurrence of megadroughts during the MedievalClimate Anomaly (MCA—850–1299 C.E.) is atypical:the five most severe and persistent droughts since 850C.E. occurred during this short time period (Coatset al 2015a). The atmosphere-ocean dynamics duringthe MCA therefore have long been a focus of scientific

research, with variability in the Pacific and AtlanticOceans having been invoked to explain the MCAhydroclimate conditions (e.g. Graham et al 2007,Herweijer et al 2007, Seager et al 2007, 2008, Fenget al 2008, Oglesby et al 2011). Nevertheless, muchremains to be understood about the MCA, particularlywith regard to the relative influence of forced andinternal variability in creating the Pacific and AtlanticOcean conditions underlying MCA hydroclimate, thecharacteristics of these ocean conditions including theirpersistence, and the likelihood of such ocean conditionsoccurring—including in the future.

Herein we use a tree-ring derived estimate of AWhydroclimate variability (the North American DroughtAtlas (NADA)—e.g. Cook et al 2004, 2007) in a time-series modeling framework to better understand theclustering of megadroughts during theMCA. Two fun-damental questions are addressed:

OPEN ACCESS

RECEIVED

28 January 2016

REVISED

13May 2016

ACCEPTED FOR PUBLICATION

27 June 2016

PUBLISHED

19 July 2016

Original content from thisworkmay be used underthe terms of the CreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2016 IOPPublishing Ltd

(1)Given the characteristics of the last 1156 years ofhydroclimate variability in the AW (850–2005C.E.), what is the likelihood of the five most severeand persistent droughts falling within an MCA-length (450 year) period?

(2)Are certain characteristics of the last 1156 years ofhydroclimate variability important for producingclustered megadroughts during theMCA and whatdoes this suggest for the dynamics underlyinghydroclimate conditions during this period?

By answering these questions we seek to betterunderstand whether several consecutive centuries ofheightened megadrought occurrence during the MCAcould arise from our current best estimates of internalvariability within the climate system or whether itrequires external forcing (e.g. solar and volcanic) and,if not, what modes of climate variability can generatemulticentury clustering of megadroughts. Togetherthis information improves understanding of futuremegadrought risk over the AW.

Data andmethods

Reconstructed Palmer drought severity index (PDSI)data are from an updated version of the tree-ringderived NADA version 2a, with improved spatialcoverage and resolution, the full details of which canbe found in Cook et al (2014a). The data arereconstructed on a 0.5°×0.5° latitude–longitude gridof JJA average PDSI values for theUnited States, as wellas parts of Canada andNorthernMexico.

The AW is defined as the region bounded by 125°W–105°W, 25°N–42.5°N. This definition is consistentwith previous research on megadroughts (e.g. Coatset al 2013, 2015a) and variability within the AW is lar-gely homogenous with an average correlation coeffi-cient of 0.74 between the AW averaged timeseries fromthe NADA and the individual NADA grid points overtheAW for the period 850–2005C.E. TheNADA is thusaveraged over the AW to create a single timeseries ofhydroclimate variability for the period 850–2005 C.E.thatwill be analyzed herein. Themost severe andpersis-tent droughts within this timeseries are identified andranked using the two start, two end (2S2E) and cumula-tive drought severity ranking criteria of Coats et al(2013). These criteria define a drought as commencingafter two consecutive years of negative PDSI and con-tinuing until two consecutive years of positive PDSI,

with the droughts then ranked by summing the PDSIfrom the first to the last year of each identified droughtfeature. For the analyses herein, focus will be restrictedto the five highest-ranking (or most severe and persis-tent) droughts. The cumulative drought severity rank-ing was chosen over a purely length-based ranking inorder to incorporate both the severity andpersistence ofeach drought. The 2S2E and cumulative drought sever-itymethodwill hereinafter be referred to as the droughtidentificationmetric.

All surrogate timeseries are produced using phaserandomization (e.g. Schreiber and Schmitz 2000),which takes the continuum of variability described inthe frequency domain (power spectral density andphase as a function of frequency) and inverts thisinformation into the time domain after randomizingthe phase. In all cases, the power spectral density usedfor producing surrogate timeseries by phase randomi-zation will be estimated using the Thomson’s multi-taper method (Thomson 1982). The outputs of phaserandomization are timeseries with the time historyrandomized and the magnitude, variance and auto-correlation properties of the prescribed continuum ofvariability preserved.

To assess the impact of precipitation and temper-ature changes on PDSI we also calculate PDSI for 500year control simulations from the Coupled ModelIntercomparison Project Phase 5 (CMIP5-Tayloret al 2012; table 1). Three models are used to test if thecharacteristics of individual models alter the impact ofprecipitation and temperature on PDSI; the specificmodels were chosen to represent a range of character-istics of low-frequency hydroclimate variability (e.g.Coats et al 2015a). Herein we use the Penman-Mon-teith (PM—Penman 1948) formulation of PDSI (a fulltreatment of the PDSI formulation can be found inCook et al 2014b). At each grid point on an even2.5°×2.5° grid, PDSI was calculated and then stan-dardized against the full 500 year period of the simula-tions. The PDSI was averaged over June–July–August(JJA) to produce a single average for each year; herein-after anymention ofmodel PDSI will be with regard tothe JJA average values.

Results

Assessment of AWhydroclimate variability(850–2005C.E.)Figure 1 plots the NADA averaged over the AW for theperiod 850–2005 C.E (hereinafter the NADA AW

Table 1.Model information for the analyzedCMIP5 simulations.

Modeling center Institute ID Model name

National Center for Atmospheric Research NCAR CCSM4

Institute Pierre-Simon Laplace IPSL IPSL-CM5A-LR

JapanAgency forMarine-Earth Science andTechnology, Atmosphere andOceanResearch Institute

(TheUniversity of Tokyo), andNational Institute for Environmental Studies

MIROC MIROC-ESM

2

Environ. Res. Lett. 11 (2016) 074025

timeseries). Throughout figure 1, the year 1299 C.E. isindicated with a vertical black line that marks the endof the MCA as defined herein (850–1299 C.E.).Although there is no absolute consensus as to whatconstitutes the MCA, 850–1299 C.E. approximatelyencompasses the range of MCA definitions in theclimate literature (e.g. Jansen et al 2007). As in Coatset al (2015a), the drought identification metric wasused to identify the five highest-ranking droughts(methods) in the NADA AW timeseries (hereinaftermegadroughts) and all five features fall within theMCA. Apart from this, the NADA AW timeseries infigure 1 (top panel) exhibits no obvious differencesbetween the MCA and the 1300–2005 C.E. period(hereinafter the little ice age/modern period—LIA/Mod). Nevertheless, to explicitly define the character-istics of the MCA relative to the LIA/Mod period,figure 1 also plots the time history of the autocorrela-tion function (ACF), the scaling exponent of spectraldensity (beta), variance andmean of AWhydroclimateover the C.E. The latter three characteristics arecalculated by sliding windows of varying lengths, from50 to 500 years, across the NADA AW timeseries. Betais a measure of the proportion of variance in high and

low frequencies, with positive beta values indicatingthat low frequencies contribute more to the varianceof the underlying timeseries than high frequencies;negative beta values indicate the opposite (e.g. Huy-bers andCurry 2006, Ault et al 2013, 2014).

There is no consistent autocorrelation structure inthe NADAAW timeseries across the range of lags con-sidered in the ACF panel of figure 1. There is, however,a vertical line of enhanced autocorrelation at the ter-mination of the MCA that may be indicative of a shiftin the mean hydroclimate conditions between the twoperiods (see the final paragraph of this section for fur-ther discussion).

For short time windows (50-to-100 years), thebeta, mean and variance exhibit a large range in valuesduring the C.E. For beta, these shifts are related tosecular trends over the short time windows. The beha-vior of the variance and mean are less clear, but againappear to be a function of the short time windows forwhich shifts in the mean state of the NADA AW time-series are possible.

The longer time windows show no change in betaover the C.E. and, as such, the MCA to LIA/Modtransition is not likely to be explained by a change in

Figure 1. (top panel)The timeseries of PDSI from theNADA averaged over theAW (125°W–105°W, 25°N–42.5°N) for the period850–2005 C.E. (NADAAWtimeseries). Internannual variability is plotted as the black bar plot and the 20 year low-pass filtered data isplotted as the blue line. The timing of thefivemost severe and persistent droughts (highest-ranking) as identified using the droughtidentificationmetric (methods) is indicated by the gray shaded regions. (second panel from top)The autocorrelation function for theNADAAWtimeseries. This is calculated out to 50 years by sliding a 50 year window across theNADAAWtimeseries (only thefirst 20years, however, are plotted). (bottom three panels)The variance,mean and beta calculated for slidingwindows varying in length from50 to 500 years. Beta is calculated using ordinary least squares in log power-log frequency space on spectra estimated using Thomson’smultitapermethod (Thomson 1982). In all panels theMCA is defined as the time before the black vertical line (850–1299 C.E.).

3

Environ. Res. Lett. 11 (2016) 074025

the spectral properties of the two periods. The var-iance, likewise, appears to be stationary across theNADA AW timeseries, although a weak shift fromlower variance in the MCA to higher variance in theLIA/Mod is apparent. This weak variance shift canpotentially be explained by the details of the recon-struction itself, as fewer tree-ring records are availablefurther back in time in theNADA.

The mean hydroclimate conditions are also drierduring the MCA as compared to the LIA/Mod,although the magnitude of this shift in PDSI units isonly about 0.3 (the variance of the NADA AW time-series, for reference, is 3.0). To determine if the lowermean during the MCA is indicative of a mean shiftrather than a longer-term trend we use the singlebreakpoint (abrupt change) identification and sig-nificance test of Lanzante (1996). A highly significant(at the 99% level) breakpoint is identified at the year1298C.E. (just before theMCA termination of 1299C.E.), suggesting that the MCA had a drier mean climatethan the LIA/Mod and that the different mean cli-mates of these two periods were separated by anabrupt change rather than being the consequence of asteady linear trend (e.g. Rodionov 2004).

Likelihood assessment ofMCAmegadroughtclusteringExperimental setupFigure 2 plots the raw spectra of the NADA AWtimeseries, in addition to the spectra of the NADAAWtimeseries with the mean of the MCA removed. Anidealized red noise spectrum with the variance andfirst-order autoregressive coefficient matching theNADA AW timeseries is also plotted in figure 2. Theshapes of the raw spectrum of the NADA AW time-series and the spectrum of the NADAAW timereserieswith the mean of the MCA removed are largelycharacteristic of a red noise process. Nevertheless,removing the mean of the MCA has the effect ofsignificantly decreasing the power spectral density on

multi-centennial-to-millennial timescales (at the 90%significance level).

To produce surrogate timeseries, the spectra infigure 2 are inverted into the time domain using stan-dard phase randomization techniques (methods). Anensemble of 10 000 of these 1156 year artificial climatehistories (length of theNADAAW timeseries) are pro-duced and subsequently used to estimate the prob-ability of occurrence ofMCAmegadrought clustering.Differences in the characteristics of the surrogates pro-duced using the raw spectrum of the NADAAW time-series and the spectrum calculated after removing theMCA mean shift (figure 2), isolate the impact of low-frequency variability like theMCAmean shift.

The surrogates are used to test the probability that,given the characteristics of the NADA AW timeseries,the five highest-ranking droughts in a 1156 year climatehistory will fall within a 450 year period (length of theMCA). Hereinafter, MCA-like will be used to describethe clustering of the five highest-ranking droughts dur-ing a 450 year subset of a 1156 year surrogate timeseries.We also calculate the average length (drought persis-tence) and cumulative drought severity (drought sever-ity) of the five highest-ranking droughts for eachsurrogate timeseries with an MCA-like period. Theseprovide an additional constraint onwhether a surrogatetimeseries contains a period that is truly representativeof the MCA, with droughts that are not only clusteredbut also characteristic of the severity and persistence ofthe actualMCAmegadroughts.

Experimental results10%percent of surrogates based on the raw spectrumofthe NADA AW timeseries contain an MCA-like period(hereinafter percent occurrence). In the surrogatesbased on the spectrum with the MCA mean shiftremoved the percent occurrence is much lower (~threepercent) suggesting that low-frequency variability liketheMCAmean shift acts to increase the probability of asurrogate timeseries containing an MCA-like period.While this result is unsurprising, figure 3 suggests that

Figure 2.Power spectral density (spectra) of PDSI from theNADA averaged over theAW (125°W–105°W, 25°N–42.5°N) for theperiod 850–2005 C.E. are plotted (blue and purple). (Blue)Raw spectra of theNADAAWtimeseries and (purple) spectra calculatedafter removing theMCAmean shift from theNADAAWtimeseries. For comparison idealized red (red)noise spectramatching thevariance and first order autoregressive coefficient of theNADAAWtimeseries is also plotted.

4

Environ. Res. Lett. 11 (2016) 074025

without low-frequency variability like the MCA meanshift, the severity and persistence of drought neverreaches the level of the actualMCAmegadroughts (boxplots do not reach the dashed lines in figure 3). In orderto get a period that is truly representative of the MCA,therefore, the shift in themeanhydroclimate conditionsduring theMCAwas necessary.

The origin of theMCAmean shiftThe results in figure 3 suggest that low-frequencyvariability manifest in features such as the MCAmeanshift in figure 1 is necessary to get drought that ischaracteristic of the actualMCAmegadroughts duringanMCA-like period. Aswas noted in the Introduction,much remains to be understood about the dynamics ofthe MCA and particularly the dynamics underlyingmegadroughts during this period. Nevertheless,research has implicated both a centuries-long warmstate of the Atlantic Ocean via the Atlantic Multi-decadal Oscillation (AMO—e.g. Feng et al 2008,Oglesby et al 2011) and cold state of the tropical PacificOcean (e.g. Graham et al 2007, Herweijer et al 2007,Seager et al 2007, 2008) in driving drying over NorthAmerica during the MCA. In this section we will testwhether these Atlantic and/or tropical Pacific Oceanconditions can explain the −0.31 mean shift in PDSIduring this period (figure 1).

PDSI is determined by two factors: precipitationand evapotranspiration. In order to have a shift in themean of PDSI during the MCA, one or both of thesefactors must have changed relative to the 1931–1990

C.E. interval over which the NADA PDSI was cali-brated and standardized. Are the canonical impacts ofa warm AMO or cold tropical Pacific sufficient to pro-duce such changes? To address this the canonicalimpact of a warm AMO and cold tropical Pacific onboth precipitation and temperature are defined usingthe instrumental record. The specific methodologywill be outlined below, but we will briefly outline twoassumptions of this test. Firstly, PDSI is a function ofmore than just precipitation and temperature, butonly changes in these two variables will be analyzed.Secondly, we will consider any precipitation or temp-erature changes to be uniform in time and acrossthe AW.

To test the impact of different precipitation andtemperature changes on PDSI we superimpose uni-form precipitation and temperature changes onmodeloutput from three control simulations (i.e. shifting themean of the precipitation and temperaturemodel out-put while keeping the variability unaltered). In eachcase, we calculate the PDSI using a PM frameworkwith all variables besides precipitation and temper-ature being the unaltered model output. For eachcombination of precipitation and temperature chan-ges we will calculate the mean PDSI over the AW forthe first 450 years (the length of the MCA) of the con-trol simulations. Results are plotted in figure 4, withgood agreement across the three models. Unsurpris-ingly, as precipitation decreases the mean PDSIdecreases and a 0.01 mm d–1 change in precipitation isassociated with an approximately 0.1 mean shift in the

Figure 3.Range in the (left panel) average cumulative drought severity and (right panel) average drought length of thefive highest-ranking droughts for surrogate timeseries that areMCA-like (defined as having thefive highest-ranking droughts fall within acontinuous 450 year period—the length of theMCA). The surrogates for theNADAAWtimeseries are produced using spectraestimatedwith themean of theMCA removed (MCAMeanRemoved) and using the raw data (Raw). All results show the full datarange, with the box plots additionally showing themedian, interquartile, and 95th and 5th percentile ranges. In both cases the value forthe five highest-ranking droughts (themegadroughts) in the actualNADAAW timeseries is plotted as the dashed black line.

5

Environ. Res. Lett. 11 (2016) 074025

PDSI. As temperature decreases the mean PDSIincreases, although at a slower rate than that associatedwith precipitation (a 0.05° C change in temperature isassociated with less than a 0.05 mean shift in thePDSI). This slower rate is consistent with PDSI pre-dominantly reflecting precipitation variability over theAW (e.g. St. George et al 2010) and the opposite sign ofthe PDSI mean shift (relative to precipitation) is con-sistent with decreased temperature leading to adecreased vapor pressure deficit (and an associateddecrease in evapotranspiration), following the Clau-sius–Clapeyron relationship acting on the saturationvapor pressure (e.g. Cook et al 2014b, Scheff andFrierson 2015).

To specifically test the potential that a persistentlywarm AMO or cold tropical Pacific produced theMCA mean shift in the NADA AW timeseries(figure 1), we calculate the canonical precipitation andtemperature impacts of these ocean states. To do so,we composite detrended (to prevent aliasing a climatechange signal) annual (January–December) precipita-tion from theGlobal PrecipitationClimatology Center(Becker et al 2013) and temperature from the NASAGoddard Global Surface Temperature Analysis (Han-sen et al 2010) for all years that are in the top third ofobserved AMO values (calculated following Enfieldet al 2001) and bottom third of observed Niño3.4values (a standard metric of the state of the tropicalPacific Ocean). Both the AMO and Niño3.4 values arecalculated from the NOAA extended reconstructedSST dataset (Smith and Reynolds 2003) over the per-iod 1901–2005 C.E. The temperature and precipita-tion composites are then averaged over the AW toproduce the canonical impacts over the region. Forprecipitation this impact is −0.036 mm d–1 (a 3%decrease in annual precipitation) for awarmAMOand−0.064 mm d–1 (a 6% decrease) for a cold tropicalPacific and for temperature it is 0.17 °C and−0.06 °C,respectively. The composite across models in figure 4suggests that a warm AMO would dry the AW slightlybeyond that of the MCA mean shift (of −0.31,figure 1). Given that temperatures over the AW were

likely cooler during the MCA than implied by thecanonical impacts of a warmAMO—because of coolerthan modern Northern Hemisphere temperatures(Moberg et al 2005, Hegerl et al 2007,Mann et al 2009)—themean PDSI values in figure 4 should be even clo-ser to the MCA mean shift. By contrast, the canonicalimpact of a cold tropical Pacific would dry the AW farin excess of the MCA mean shift. Importantly, how-ever, weaker cooling in the tropical Pacific—that is tosay, much smaller in magnitude than the cold tropicalPacific analyzed herein—would produce drying that iscloser to the MCA mean shift. A linear regression ofthe Niño3.4 index on precipitation over the AW, forinstance, suggests that a Niño3.4 index of −0.3 °C(32% of the average Niño3.4 index for a cold tropicalPacific) produces drying approximately equal to thatof theMCAmean shift.

Discussion of climatic implications andconclusions

Aperiodwith clustering of severe andpersistent droughtlike theMCA is not a highly probable event, neverthelesssuch a feature occurs in approximately 10%of surrogatetimeseries that were constructed to have the samespectral characteristics as the NADA AW timeseries.This suggests that while megadrought clustering duringthe MCA was improbable, with a recurrence interval ofapproximately 11 000 years, it is a naturally occurringphenomenon that is expected to occur over the AW.Importantly, a shift in the mean hydroclimate condi-tions during the MCA (850–1299 C.E.) is found to be acritical characteristic of the NADA AW timeseries.Without low frequency variability capable of generatingsuch a mean shift, the probability of drought clusteringlike theMCAdrops to approximately three percent and,more importantly, drought never reaches the severity orpersistence of the actualMCAmegadroughts.

PDSI over the AW calculated after superimposingthe precipitation and temperature impacts of a weakcooling in the tropical Pacific or the canonical impacts

Figure 4.MeanPDSI over theAW for the first 450 years (length of theMCA) of theCCSM, IPSL andMIROCcontrol simulationsfrom theCMIP5. Each square represents the 450 yearmean value over the AWof PDSI calculated after adding a uniform temperatureand precipitation change to themodel output over the full 450 years. The panel labeled ‘composite’ is an average over the threemodels, however only 450 yearmean PDSI valueswithin ten percent of the actualMCAmean PDSI (−0.31 PDSI—from theAWPDSItimeseries) are plotted.

6

Environ. Res. Lett. 11 (2016) 074025

of a warm AMO can largely reproduce the shift inmean hydroclimate conditions during the MCA. Acenturies-long persistence of one of these ocean states,therefore, may have been a necessary condition for theclustering of severe and persistent drought during theMCA. We hypothesize that there was a persistentlywarm AMO during the MCA (as proposed by Fenget al 2008, Oglesby et al 2011), in particular, becausethe AMO is associated with much longer timescales ofvariability (with a quasi-periodic cycle of 50 to 70 years—e.g. Mann and Park 1994) than the tropical PacificOcean, which is dominated by seasonal-to-decadaltimescales. For instance, in an additional analysisusing an ensemble of 10 000 1156 year surrogate time-series with the same power law relationship (beta) asthe observed AMO, 41% of the surrogate climate his-tories contain a 450 year period (the length of theMCA) where at least 75% of years are in the top thirdof observed AMO values. Perhaps unsurprisingly,given the long timescales of variability inherent to theAMO, this suggests that the observed characteristics ofthismode of variability—whether forced or internal—are sufficient to explain a shift towards a persistentlywarm state during theMCA.

Importantly, the AMO is only being implicated inthe mean hydroclimate shift during the MCA (e.g.figure 1) and the timing of individual MCA mega-droughts may still be related to decadal periods of coldconditions in the tropical Pacific Ocean, which occurfrequently in observations and can produce severedrying over the AW (e.g. McCabe et al 2004). Becauseit is often thought that the AMO is an internal mode ofvariability, and there is little doubt that decadal varia-bility of the tropical Pacific Ocean is, it follows thatMCA megadroughts could have arisen via purelyinternal climate dynamics despite widespread con-jecture that they were exogenously forced (e.g. Gra-ham et al 2007,Mann et al 2009). The results presentedhere can neither prove nor disprove either hypothesisbut emphasize the need to better characterize theocean states during the common era.

Such an understanding is particularly importantwhen considering future drought conditions over theAW. If a persistently warm or cold AMO occurs as aresponse to anthropogenic greenhouse gas forcing, thiswould have consequences for the probability of MCA-like drought occurring in the future. Themodels used forfuture projections, however, are unlikely to simulate theworst-case scenario of a persistently warm AMO anddecadal periods of cold conditions in the tropical Pacificbecause few models have been shown to simulate mega-droughts consistently associated with tropical PacificOcean (Coats et al 2015a) andmostmodels lack a realisticAMO and associated hydroclimate impacts over the AW(Ting et al 2011, Coats et al 2015b). These model defi-ciencies can perhaps explain the inability of the CMIP5models to produceMCA-like drought clustering, despitesimulating individual megadroughts that are as severeand persistent as those in the paleoclimate record (Coats

et al 2015a). In any case, efforts to characterize thedynamics of past hydroclimate change are criticalbecause they suggest that theCMIP5 ensemble is unlikelyto capture the full range of potential future hydroclimatestates (as also noted in Ault et al 2013, 2014; although theopposite has been implied—Frank et al 2013). Furtherattempts to characterizemodel projections in the contextof long records of climate variability will therefore con-tinue to improve the basis for accurate predictions of thefuture climate state.

Acknowledgments

This work was supported in part by NSF grants AGS-1243204 and AGS-1401400. We acknowledge theWorld Climate Research Programme’s WorkingGroup on Coupled Modelling, which is responsiblefor CMIP, and we thank the climate modeling groups(listed in table 1) for producing and making availabletheir model output. For CMIP, the U.S. Departmentof Energy’s Program for ClimateModel Diagnosis andIntercomparison provides coordinating support andled development of software infrastructure in partner-ship with the Global Organization for Earth SystemScience Portals. LDEO contribution 8035. We thanktwo anonymous reviewers for comments thatimproved the quality of thismanuscript.

References

Ault TR, Cole J E,Overpeck J T, PedersonGT, St. George S,Otto-Bliesner B,Woodhouse CA andDeser C 2013Thecontinuumof hydroclimate variability inwesternNorthAmerica during the lastmillennium J. Clim. 26 5863–78

Ault T R, Cole J E, Overpeck J T, PedersonGT andMekoDM2014 Assessing the risk of persistent drought using climatemodel simulations and paleoclimate data J. Clim. 277529–49

Becker A, Finger P,Meyer-Christoffer A, Rudolf B, SchammK,SchneiderU andZieseM2013Adescription of the globalland-surface precipitation data products of the globalprecipitation climatology centre with sample applicationsincluding centennial (trend) analysis from1901-presentEarth Syst. Sci. Data 5 71–99

Coats S, CookB I, Smerdon J E and Seager R 2015bNorthAmericanpan-continental drought inmodel simulations of the lastmillennium J. Clim. 28 2025–43

Coats S, Smerdon J E, CookB I and Seager R 2015aAre simulatedmegadroughts in theNorthAmerican southwest forced?J. Clim. 28 124–42

Coats S, Smerdon J E, Seager R, CookB I andGonzález-Rouco J F2013Megadroughts in southwestNorth America inmillennium-length ECHO-G simulations and theircomparison to proxy drought reconstructions J. Clim. 267635–49

CookB I, Smerdon J E, Seager R andCook ER 2014a Pan-continental droughts inNorth America over the lastmillennium J. Clim. 27 383–97

CookB I, Smerdon J E, Seager R andCoats S 2014bGlobal warmingand 21st century dryingClim.Dyn. 43 2607–27

Cook ER, Seager R, CaneMAand StahleDW2007NorthAmerican drought: reconstructions, causes, andconsequences Earth Sci. Rev. 81 93–134

Cook ER, Seager R,HeimRR Jr, Vose R S,Herweijer C andWoodhouseC 2010Megadroughts inNorthAmerica:

7

Environ. Res. Lett. 11 (2016) 074025

placing IPCCprojections of hydroclimatic change in a long-termpalaeoclimate context J. Quat. Sci. 25 48–61

Cook ER,Woodhouse CA, EakinCM,MekoDMand StahleDW2004 Long-term aridity changes in thewesternUnited StatesScience 306 1015–8

EnfieldDB,Mestas-NuñezAMandTrimble P J 2001TheAtlanticmultidecadal oscillation and its relation to rainfall andriverflows in the continental USGeophys. Res. Lett. 282077–80

Feng S,Oglesby R J, RoweCM, LoopeDB andHuQ2008Atlanticand Pacific SST influences onmedieval drought inNorthAmerica simulated by theCommunity AtmosphericModelJ. Geophys. Res. 113D11101

Franke J, FrankD, Raible CC, Esper J and Brönnimann S 2013Spectral biases in tree-ring climate proxiesNat. Clim. Change3 360–4

GrahamNE et al 2007Tropical Pacific -mid-latitudeteleconnections inmedieval timesClim. Change 83 241–85

Hansen J, Ruedy R, SatoMand LoK 2010Global surfacetemperature changeRev. Geophys. 48RG4004

Hegerl GC et al 2007Detection of human influence on a new,validated 1500-year temperature reconstruction J. Clim. 20650–66

Herweijer C, Seager R, Cook ER and Emile-Geay J 2007NorthAmericanDroughts of the lastmillennium from a griddednetwork of tree-ring data J. Clim. 20 1353–76

Huybers P andCurryW2006 Links between annual,Milankovitchand continuum temperature variabilityNature 441 329–32

Jansen E et al 2007 PalaeoclimateClimate Change, The PhysicalScience Basis. Contribution ofWorkingGroup I to the fourthassessment report of the Intergovernmental Panel on ClimateChange ed S Solomon et al (Cambridge: CambridgeUniversity Press) pp 433–97

Lanzante J R 1996Resistant, robust and non-parametric techniquesfor the analysis of climate data: theory and examples,including applications to historical radiosonde station dataInt. J. Climatol. 16 1197–226

MannME and Park J 1994Global scalemodes of surfacetemperature variability on interannual to century time scalesJ. Geophys. Res. 99 25819–33

MannME et al 2009Global signatures and dynamical origins of theLittle Ice Age andMedieval Climate Anomaly Science 3261256–60

McCabeG J, PaleckiMA andBetancourt J L 2004 Pacific andAtlantic ocean influences onmultidecadal drought frequencyin theUnited StatesProc. Natl Acad. Sci. USA 101 4136–41

Moberg A et al 2005Highly variableNorthernHemispheretemperatures reconstructed from low-and high-resolutionproxy dataNature 433 613–7

Oglesby J R, Feng S,HuQ andRoweC 2011Medieval drought inNorthAmerica: the role of the AtlanticmultidecadaloscillationPAGESNews vol 19,No. 1 (Bern, Switzerland:PAGES International ProjectOffice), pp 18–9

PenmanHL 1948Natural evaporation fromopenwater, bare soiland grassProc. R. Soc.A120–45

Rodionov SN 2004A sequential algorithm for testing climateregime shiftsGeophys. Res. Lett. 31 L09204

Scheff J and FriersonDM2015Terrestrial aridity and its response togreenhousewarming across CMIP5 climatemodels J. Clim.28 5583–600

Schrieber T and Schmitz A 2000 Surrogate time series PhysicaD 142346–82

Seager R, BurgmanR, Kushnir Y, Clement A, Cook E,NaikN andMiller J 2008Tropical Pacific forcing ofNorth Americanmedievalmegadroughts: testing the concept with anatmospheremodel forced by coral-reconstructed SSTsJ. Clim. 21 6175–90

Seager R,GrahamN,Herweijer C, GordonAL, Kushnir Y andCook E 2007 Blueprints formedieval hydroclimateQuat. Sci.Rev. 26 2322–36

Smith TMandReynolds RW2003 Extended reconstruction ofglobal sea surface temperatures based onCOADSdata(1854–1997) J. Clim. 16 1495–510

St. George S,MekoDMandCook ER 2010The seasonality ofprecipitation signals encodedwithin theNorthAmericanDrought AtlasHolocene 20 983–8

StahleDW,Cook ER,CleavelandMK,TherrellMD,MekoDM,Grissino-MayerHD,Watson E and LuckmanBH2000Tree-ring data document 16th centurymegadrought overNorthAmericaEos Trans. AGU 81 121–5

StahleDW, Fye FK andCook ER 2007Tree-ring reconstructedmegadroughts overNorth America since AD1300Clim.Change 83 133–49

Stine S 1994 Extreme and persistent drought inCalifornia andPatagonia duringmedieval timeNature 369 546–9

Taylor K E, Stouffer R J andMeehl GA 2012Anoverview of CMIP5and the experiment designBull. Am.Meteorol. Soc. 93 485–98

ThomsonD J 1982 Spectrum estimation and harmonic analysisProc. IEEE 70 1055–96

TingM,Kushnir Y, Seager R and Li C 2011Robust features ofAtlanticmulti‐decadal variability and its climate impactsGeophys. Res. Lett. 17 L17705

8

Environ. Res. Lett. 11 (2016) 074025