Six Hundred Years of South American Tree
Six Hundred Years of South American Tree
Six Hundred Years of South American Tree
Edited by James C. Zachos, University of California, Santa Cruz, CA, and approved June 1, 2020 (received for review February 18, 2020)
South American (SA) societies are highly vulnerable to droughts Argentina during the drought years 1968 to 1969, 1976 to 1977,
and pluvials, but lack of long-term climate observations severely 1996 to 1997, and 2010 to 2019 due to a deficit in snow accu-
limits our understanding of the global processes driving climatic mulation in the Andes. Farmers in western Argentina reported
variability in the region. The number and quality of SA climate- between 35% and 50% crop losses as a consequence of water
sensitive tree ring chronologies have significantly increased in re- supply reduction in 1968/69 (1). This specific event, known as
cent decades, now providing a robust network of 286 records for the Great Drought of 1968, also had serious impacts in central
characterizing hydroclimate variability since 1400 CE. We combine Chile. Cereal and vegetable production dropped by 65%, ir-
this network with a self-calibrated Palmer Drought Severity Index rigation areas contracted by 40%, and livestock numbers de-
(scPDSI) dataset to derive the South American Drought Atlas creased by 45% across the country, leading to the loss of
(SADA) over the continent south of 12°S. The gridded annual re- 225,000 agricultural jobs (2). Presently, this region is experi-
construction of austral summer scPDSI is the most spatially com- encing the most severe decadal drought identified during the
plete estimate of SA hydroclimate to date, and well matches past last millennium (3). Socioeconomic disasters such as those
historical dry/wet events. Relating the SADA to the Australia–New associated with the 1968 drought highlight the acute vulner-
Zealand Drought Atlas, sea surface temperatures and atmospheric ability of SA to extreme droughts, motivating improved un-
pressure fields, we determine that the El Niño–Southern Oscilla- derstanding of the occurrence, duration, and spatial extent of
tion (ENSO) and the Southern Annular Mode (SAM) are strongly
associated with spatially extended droughts and pluvials over the
Author contributions: M.S.M., E.R.C., J.B., D.A.C., R.V., C.L., B.H.L., and J.A.B. designed
SADA domain during the past several centuries. SADA also ex- research; M.S.M. performed research; M.S.M., E.R.C., J.B., A.G.-R., F.C., and V.M. contrib-
hibits more extended severe droughts and extreme pluvials since uted new reagents/analytic tools; M.S.M., E.R.C., J.B., D.A.C., R.V., C.L., A.M.S., M.E.F.,
the mid-20th century. Extensive droughts are consistent with the A.G.-R., F.C., V.M., F.R., M.R.P., J.E.S., L.O.B., M.H.M., A.A.M., M.R.-B., C.A., L.L., D.L.,
I.H., P.D.J., A.P.W., G.V., D.A., I.A.-B., E.M., F.F., T.M., and E.C. analyzed data; M.S.M. wrote
observed 20th-century trend toward positive SAM anomalies con-
the paper; E.R.C., J.B., D.A.C., R.V., C.L., A.M.S., M.E.F., A.G.-R., F.C., V.M., J.C.A., A.L.,
comitant with the weakening of midlatitude Westerlies, while I.A.M., F.R., M.R.P., J.E.S., L.O.B., R.U.-J., M.R.-C., B.H.L., P.D.J., A.P.W., and J.A.B. reviewed
low-level moisture transport intensified by global warming has and edited versions of the paper; R.V., D.A.C., C.L., M.E.F., J.C.A., A.L., I.A.M., M.H.M.,
favored extreme rainfall across the subtropics. The SADA thus R.U.-J., M.R.-C., A.A.M., M.R.-B., C.A., L.L., G.V., D.A., I.A.-B., F.F., T.M., E.C., and J.A.B.
provided tree ring chronologies and interpretation; M.S.M., J.B., A.M.S., F.C., L.O.B.,
provides a long-term context for observed hydroclimatic changes
M.R.-C., and E.M., compiled the instrumental database; and D.L., I.H., and P.D.J. produced
and for 21st-century Intergovernmental Panel on Climate Change the gridded instrumental database.
(IPCC) projections that suggest SA will experience more frequent/ The authors declare no competing interest.
severe droughts and rainfall events as a consequence of increasing
This article is a PNAS Direct Submission.
greenhouse gas emissions.
This open access article is distributed under Creative Commons Attribution-NonCommercial-
NoDerivatives License 4.0 (CC BY-NC-ND).
| |
drought atlas palaeoclimate reconstruction extreme hydroclimate
Data deposition: Instrumental and reconstructed scPDSIs have been deposited at the
| |
events South America hydroclimate Southern Hemisphere climate modes Center for Climate and Resilience Research (CR)2, http://www.cr2.cl/datos-dendro-sada/.
1
To whom correspondence may be addressed. Email: mmorales@mendoza-conicet.gob.ar.
P roductive economic and social activities in South America This article contains supporting information online at https://www.pnas.org/lookup/suppl/
Downloaded by guest on July 23, 2020
EARTH, ATMOSPHERIC,
projections for the areas adjacent to the Andes present large drought in the subtropical Andes (12). The drying trend along
uncertainties due to the limited capacity of global climate the western-central coast of SA during the past decades is also
models to simulate processes at regional scales in mountain- consistent with the expansion of the Hadley cell and the pole-
ous regions (4). ward shift of the westerlies (12–14). During cold ENSO phases,
A B C
Altiplano
Central
Chile La Plata
basin
Start year
< 1300
1300 - 1399
1400 - 1499
1500 - 1599
1600 - 1699
1700 - 1799
1800 - 1899
0 230 460 Km
cl 95% cl 95%
from simulated events
Departures of Actual
0.0 0.0
0.0
Dry years
-0.5 -0.5
-0.5
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Lag (yrs) Lag (yrs) Lag (yrs)
Fig. 1. SADA domain and verifications based on instrumental climate and historical documents. (A) Map of the DJF scPDSI target field (small orange
dots mark grid-point centers) and the network of 286 tree ring chronologies used for reconstruction (circles colored to indicate start year). (B)
Calibration-period regression coefficient of determination (CRSQ); the white region over northern Chile indicates where the reconstruction was not
performed. (C ) Calibration period leave-one-out cross-validation reduction of error (CVRE). (D–F ) Superposed epoch analyses (SEAs) for reconstructed
scPDSI (red rectangles in map) during the dry/wet events recorded by historical documents from (D) Potosí, Bolivia (1585 to 1807), (E ) central Chile
(1530 to 2000), and (F ) Santa Fe city (1585 to 1815; La Plata basin). The red/blue bars represent scPDSI departure from normal conditions for a 9-y
Downloaded by guest on July 23, 2020
window (t − 4 to t + 4) based on 1,000 Monte Carlo simulations for the dry/wet historical events, respectively. The short dashed lines represent the 95%
confidence limits.
Morales et al. PNAS | July 21, 2020 | vol. 117 | no. 29 | 16817
the poleward shifts of the westerlies were enhanced by the the calibration. Given the relatively short instrumental period used
concomitant SAM positive phase (15, 16). Hence, the expansion for calibration (1951 to 2000) and the declining number of the in-
of the Hadley cell descending branch concurrent with the strumental series used in the scPDSI grid prior to 1951 (SI Ap-
southward shift of the westerlies has increased the frequency of pendix, Fig. S1), the reconstructions were primarily validated using a
dry events in the subtropical Andes, the temperate and cold leave-one-out cross-validation procedure (24, 25) (SI Appendix,
regions of SA (12, 13). section 6, Fig. 1B, and SI Appendix, Fig. S4 A–C). The fraction of
Despite all these alarming trends in SA’s hydroclimate, the variance explained by regression in the calibration period (CRSQ
brevity of available instrumental climate records (usually less [1951 to 2000]) is above 20% for almost the entire SADA domain
than 60 y) can only provide a limited view of extremes in natural and above 40% for more than 35% of the domain (Fig. 1B). The
hydroclimate variability. A long-term and large-scale perspective cross-validation reduction of error (CVRE) similarly indicates that
is necessary to document the historical range and sequence of over 20% of the variance is explained by the tree ring data in more
hydroclimate variations in SA, their connections to large-scale than 75% of SADA domain (Fig. 1C).
climate modes and their interactions (e.g., ENSO and the SAM), Validation results are similar to those recorded for other
and the impacts of external climate forcings. In many regions of drought atlases, such as the European Old World Drought Atlas
SA, climate proxies such as tree rings, ice cores, sediments, (23), and suggest that the overall SADA is skillful. To further
speleothems, and historical archives have been used successfully validate the SADA results, however, we compare our recon-
to reconstruct hydroclimatic variations over centuries to mil- structions to independent multicentury historical archives (18) in
lennia (17–19). However, the local geographic nature of such three different regions of SA: the Altiplano, central Chile, and
records reduces their applicability over large spatial regions, the La Plata basin (SI Appendix, section 8 and Fig. S6). For this
making it difficult to achieve the broad and detailed coverage purpose, a superposed epoch analysis (SEA) was conducted
needed to resolve the spatiotemporal complexity of SA hydro- (Materials and Methods), showing significant below/above scPDSI
climatic variability. We address these limitations by bringing values are associated with the dry/wet extreme years recon-
together a network of observational climate data and tree ring structed by documentary data for the three regions (Fig. 1 D–F).
records to develop the South American Drought Atlas (SADA), The highly significant phasing between the entirely independent
a spatiotemporal reconstruction of austral summer hydroclimate scPDSI reconstruction and historical records add considerable
between 12° and 56°S latitude in SA, which allows us to char- confidence to both proxies as reliable sources of information for
acterize the spatiotemporal variability of extreme droughts and past hydroclimate variations across SA during past centuries.
pluvials over the last 600 y. We also use the spatial covariance
between the SADA and the Australian New Zealand Drought Temporal Patterns of Extreme Events. There is considerable
Atlas (ANZDA) (20) to identify and reconstruct the spatial year-to-year variability in the estimates of the wet/dry conditions
fingerprints of the major climate drivers affecting hydroclimate across the SADA domain (Fig. 2 A and B), while the average
variations over the SADA–ANZDA domain, which we then use to scPDSI for all 2,715 grid cells shows large decadal to centennial
help diagnose the coupled ENSO/SAM events as drivers of extreme variability; the SADA domain has moved between wetter (e.g.,
droughts and pluvials over the last half millennium in SA. early 19th century) and drier (e.g., mid-18th century) climates
repeatedly over the past 600 y (Fig. 2B).
Results A temporally varying return-time analysis of the spatial extent
The SADA is a tree ring-based spatiotemporal reconstruction of of severe droughts/pluvials (scPDSI less than −2/greater than
austral summer (December, January, February [DJF]) droughts +2, respectively) reveals the frequency of widespread severe
and pluvials over a large portion of SA (12° to 56°S) spanning the droughts over the past six centuries was nonstationary (Fig. 2C).
last 600 y. The reconstruction has been produced on a 2,715 point There has been a steady increase in the frequency of widespread
regular latitude–longitude grid with 0.5° resolution (Fig. 1A), and droughts since about 1930, with the highest frequency (one event
overall reflects contemporaneous summer soil moisture conditions, per ∼10 y) since the 1960s, while the return time of severe
as well as those accumulated over previous seasons. The spatial widespread pluvials has remained relatively steady at 16 to 34 y
extent and duration of the SADA make it the most spatially complete, over the period of study (Fig. 2C). A return time analysis of the
highest-resolution, annually-resolved hydroclimate reconstruction that spatial extent of both droughts and pluvials identifies the highest
has specifically targeted SA. frequency of severe widespread events every ∼5 y post-1960,
The 286 tree ring chronology network used for reconstruction slightly higher than the second most frequent return period (∼6
is shown in Fig. 1A. Compared to the self-calibrated Palmer y) recorded during the 17th century (gray line in Fig. 2C). A
Drought Severity Index (scPDSI) grid, its coverage is not spatially return time analysis of the average scPDSI value for the whole
uniform across SA, mainly due to the dearth of long-lived trees in SADA domain (Fig. 2D) estimates that extreme pluvial events
lowlands and deserts, the lack of suitable species for dendrochronol- ocurrence were the most frequent since the second half of 20th
ogy, and/or the absence of chronologies from species with dendro- century (one event per ∼11 y). The recurrence interval of ex-
chronological potential. Despite these limitations, a well-calibrated treme drought during the second half of the 20th century was
and validated reconstruction of DJF scPDSI has been produced short (∼12 y) but not different from the previous 17th-century
using an ensemble-based modification of the point-by-point re- period (∼13 y). Combining the occurrence of extreme droughts
gression (PPR) method applied in previous drought atlases and pluvials, the shortest return time corresponds to the second
(20–23) (SI Appendix, sections 5 and 7 and Fig. S5). half of the 20th century with one extreme event every ∼4.5 y,
The instrumental scPDSI data cover the 1901 to 2015 period while the second shortest rate was recorded during the 17th
(SI Appendix, section 2). In contrast, the tree ring chronologies century (one event per 5.5 y). Both analyses collectively indicate
have a common end year of 2000 because of the wide variation in that widespread severe droughts and extreme pluvials over the
the dates of tree ring sampling. The 1901 to 1950 scPDSI data are SADA domain have been more frequent during the second half
of weaker quality, relative to data after 1951, due to the declining of the 20th century, relative to the rest of the SADA period.
number of precipitation records available for the scPDSI grid (SI
Appendix, Figs. S1 and S2). For this reason, the selected period for Hydroclimate Extremes and Societal Impacts. Four well-documented
the reconstruction models was 1951 to 2000 because it is based on extreme hydroclimatic events from three different regions are
the best quality instrumental data. The most common method used characterized by historical information completely independent of
Downloaded by guest on July 23, 2020
in dendroclimatology to validate a reconstruction is to compare the the SADA; note that the complete historical hydroclimatic recon-
tree ring climate estimates with instrumental data not included in structions were compared in Fig. 1 D–F and SI Appendix, Fig. S6 (SI
EARTH, ATMOSPHERIC,
margins such as the city of Santa Fe (mid-Paraná River basin, Dry (wet) conditions were also indicated in the Altiplano and
Argentina) have provided abundant documentary evidence of central-east Australia during positive (negative) DJF SSTs. The
significant flood events. Based on this information, Prieto (29) correlation coefficients between the DJF SSTs for the Niño 3.4
reconstructed 38 large flood events between 1585 and 1815. We sector (HadISST_N3.4; SI Appendix, section 10) and the scPDSI
highlight here the 1651 to 1652 and 1723 floods. The floods of leading mode and SST leading modes are 0.63 (Fig. 4C) and
1651 to 1652 destroyed more than half of the city of Santa Fe, 0.94, respectively. A similar spatial pattern emerges by applying
leading to the relocation to its current position (29). Consistent MCA to instrumental SSTs and scPDSI datasets (SI Appendix,
with this event, the 1651 SADA map (Fig. 3C) shows very wet Fig. S8A) with a leading mode of coupled variability explaining
conditions throughout the La Plata basin, gradually decreasing 57% of the total covariance.
toward the Chaco region in Argentina, southern Brazil, and The MCA was repeated using DJF geopotential height at 500
Patagonia. In 1723, the villages and cultivated fields near the mb (GPH500; SI Appendix, section 10) and reconstructed DJF
A B 2
50
% Dry and wet areas
PDSI index
25
0 0
25 −1
P95
P95 P05
Severe wet Extreme wet Severe dry Extreme dry
50 −2
C 1400 1500 1600 1700 1800 1900 2000 1400 1500 1600 1700 1800 1900 2000
D
Return time (Years)
Return time (Years)
4.5 4.5
Droughts Droughts
Pluvials Pluvials 5.5
5.5
Droughts + Pluvials Droughts + Pluvials
7.5 7.5
10.0 10.0
20.0 20.0
80.0 80.0
1400 1500 1600 1700 1800 1900 2000 1400 1500 1600 1700 1800 1900 2000
Year Year
Fig. 2. Spatial magnitude and frequency of extreme drought and pluvial events in the SADA domain. (A) Percentage of area under severely dry (scPDSI −2,
orange) and extremely dry (scPDSI −4, red) conditions, and for severely wet (scPDSI +2, light blue) and extremely wet (scPDSI +4, blue) conditions. The black
short dashed lines indicate 95% percentiles of the distribution, from which severe widespread dry/wet events were selected for the return time analysis in C.
(B) Average of SADA reconstructions over the entire study domain. The red and blue short dashed lines indicate the 5% and 95% percentiles of the dis-
tribution, respectively, from which extreme drought/wet events were selected for the return time analysis in D. (C and D) Time-varying frequency of the
occurrences of severe widespread dry/wet scPDSI events and extreme dry/wet scPDSI events, respectively, between 1400 and 2000. A kernel smoothing
Downloaded by guest on July 23, 2020
method was used with a bandwidth of 50 y (49). The shaded areas (gray, blue, and orange) represent 95% confidence intervals based on 1,000 bootstrap
simulations.
Morales et al. PNAS | July 21, 2020 | vol. 117 | no. 29 | 16819
Fig. 3. Austral summer scPDSI maps of historical extremely dry/wet events in three regions from the SADA domain. (A) The silver mine drought of Potosí,
Bolivia (1800 to 1804). (B) The central Chile drought of 1863 (1863 to 1866). (C) Santa Fe city floods (1651, 1723). The red stars indicate the geographic location
of these recorded historical events.
scPDSI from SADA and ANZDA over the period 1948 to 2000. (Fig. 4 C and D). The relationships between different paleo-
The resulting coupled spatiotemporal pattern accounts for 32% climate ENSO and SAM reconstructions and our 500-y ENSO-e
of the total cross-covariance and is consistent with the SAM (10) and SAM-e estimates are shown in SI Appendix, Figs. S9 and
spatial pattern. The SAM, which is the primary mode of tropo- S10; correlations between the reconstructions are statistically
spheric circulation variability south of 20°S (31), was correlated significant in most cases.
with the first covariance-leading mode (scPDSI, r = 0.55; The ENSO-e and SAM-e time series are negatively correlated
Fig. 4D; GPH500, r = 0.67). Variations in the SAM phases result (r = −0.60; P < 0.0001) over CE 1500 to 2000, providing a first
from atmospheric mass exchanges between the sub-Antarctic (50 indication of persistent interactions between ENSO and SAM
to 60°S) and mid-latitudes (40 to 50°S) in the SH. The positive over the past 500 y in the sense that La Niña causes a poleward
phase is associated with decreased geopotential heights over shift of the westerlies and, hence, a positive SAM. To explore the
Antarctica and strengthened and more poleward SH westerlies. influence of combinations of ENSO/SAM modes over the SA
Consequently, drier conditions occur in central Chile, Patagonia, hydroclimate during the past 500 y, we examined those years
New Zealand, and Tasmania during the positive phase (Fig. 4B). when ENSO-e was negative and SAM-e was positive, and vice
During the negative phase, opposite conditions are observed. versa. A total of 25 (26) events for each phase was identified
Similar patterns emerged when MCA was applied to the in- showing negative (positive) ENSO-e and positive (negative)
strumental GPH500 and scPDSI datasets, showing 28% of SAM-e coefficients. The SADA composite map for the 25 cou-
shared covariance between these fields (SI Appendix, Fig. S8B). pled negative ENSO-e/positive SAM-e years (Fig. 4E) shows
The leading modes of covariance were extended over the com- dominant dry conditions over central Chile, Patagonia, and
mon 1500 to 2000 period between SADA and ANZDA, and southeast SA, whereas the Altiplano and northwestern Argen-
Downloaded by guest on July 23, 2020
used as estimators of past ENSO and SAM variability in the SH tina are wet. The 26-y composite scPDSI map for positive
domain, hereafter called ENSO-e and SAM-e, respectively ENSO-e/negative SAM-e years shows wet conditions across most
4
C r = 0.63 low-level jet in SA, causing a wetting in SESA (39). However, the
1st covariance mode
ihadiSSTs_N3.4
2 wetting trend over SESA and the poleward shift of the westerlies
0 in summer are also modulated by ozone depletion over Ant-
EARTH, ATMOSPHERIC,
-2
ANZDA in the SH. The simultaneous use of the SADA and
r = 0.55 ANZDA allow us to determine the influence of tropical–
1500 1600 1700 1800 1900 2000 extratropical interactions on hydroclimate variability in the Pa-
E F cific domain of the SH over the past several centuries. Using
ENSO_e (-) ENSO_e (+) MCA (Fig. 4), we isolated the coupled patterns of maximum-
-20° -20° SAM_e (-)
SAM_e (+) shared covariance between reconstructed scPDSI from SADA
and ANZDA, with hemispheric SST and pressure field varia-
-30°
self-calibrating PDSI
-30°
2
SAM during recent decades and the consequent poleward shift Observational Climate and Tree Ring Data. The observational data used to
of the westerlies and associated storm tracks, leading to drier compute instrumental scPDSI was an ensemble of the interpolated fields of
Morales et al. PNAS | July 21, 2020 | vol. 117 | no. 29 | 16821
monthly climate observations from three datasets: 1) precipitation, tem- with scPDSI values less than −2 for severe dry and greater than 2 for severe
perature, and potential evapotranspiration data from the Climatic Research wet years. Those severe dry/wet events that exceeded the 95th percentile of
Unit Time Series (CRU) TS 4.01 (42), enhanced by the incorporation of pre- spatial extent were considered extreme spatially widespread pluvial/drought
cipitation and temperature records from the Argentinean Institute of Snow, events. Temporary changes in the occurrence-rate estimation of extremes in
Ice and Environmental Research (IANIGLA-CONICET) database (SI Appendix, intensity and spatial extent of drought/pluvial events were estimated using
section 1 and Fig. S1); 2) precipitation and air temperature dataset from the the nonparametric kernel function. This technique allows the detection of
University of Delaware (43); and 3) the precipitation dataset from Global nonmonotonic trends without imposing parametric restrictions. For this
Precipitation Climatology Centre (44). The computed monthly scPDSI data
purpose, a Gaussian kernel function was applied in order to estimate the
were seasonalized to develop average data for the austral summer season
probability of occurrence of one specific extreme event using a 60-y band-
(DJF). The DJF scPDSI reflects spring–summer soil moisture conditions from
width. To better interpret these estimates, confidence bands at the 95%
2,715 grid cells (0.5° longitude by 0.5° latitude) covering the study domain
level were obtained using 1,000 bootstrap simulations (48, 49).
(12° to 56°S; 50° to 80°W) (SI Appendix, section 2).
During the last several decades, tree ring scientists from Chile, Argentina,
and Bolivia have vastly increased tree ring sample collections in SA (17). The MCA for Reconstructed scPDSI and Climate Modes. To describe how SST and
SADA includes data from 286 tree ring chronologies (SI Appendix, section 3 geopotential height (500 mb) (SI Appendix, section 10, for dataset reference)
and Table S1), mainly concentrated on both sides of the Andes Cordillera covary with the summer scPDSI from the SADA and the ANZDA, we used
(16° to 56°S), from the Altiplano and intermontane subtropical valleys to the MCA. This method is widely used in climate research and identifies coupled
Patagonian forests at the southern tip of the continent (Fig. 1A). Addition- patterns in two data fields that share the maximum amount of covariance
ally, new collections from tropical lowlands have allowed extension of the (50). This statistical tool identifies the common signal while separating sto-
geographical coverage of tree ring records to lower latitudes. The target chastic noise from other factors. The leading modes obtained by MCA were
period (CE 1400 to 2000) for reconstruction of scPDSI is the consequence of a used as estimators of ENSO (ENSO-e) and SAM (SAM-e) variability for the
relatively high number of longer tree ring chronologies and good spatial past 500 y (Dataset S2). The resulting time series of the difference between
coverage along the Andes (SI Appendix, Fig. S3). To preserve medium fre- both climate index estimators was used to determine the anomalous neg-
quency variability due to climate, the 286 tree ring chronologies were ative/positive values by the 5th and 95th percentiles, respectively (Dataset
standardized using “signal-free” methods (45) (SI Appendix, section 4). S3). The 25 (26) negative (positive) values were associated with coupled
anomalous negative (positive) ENSO-e and positive (negative) SAM-e events.
Reconstruction Method. A modification of the well-tested nested PPR method
(22) was used to produce the SADA (SI Appendix, section 5). The version Data Deposition. Tree ring chronologies, instrumental and reconstructed
presented here is an average derived from an ensemble approach using
scPDSI (SADA) are available at the Center for Climate and Resilience Research
15-ensemble members, where each member uses a different search radius
(CR)2, http://www.cr2.cl/datos-dendro-sada/ (51). Historical hydroclimate re-
(200-, 500-, 800-, 1,100-, and 1,500-km distance) to locate the tree ring
constructions together with regional scPDSI used to validate each proxy are
chronologies for reconstructing scPDSI at each grid point based on a
presented in Dataset S1. The main leading modes used as estimators of ENSO
weighted power correlation (P = 0; P = 1; P = 2) between tree ring chro-
nologies and scPDSI (23). The 15 output model members were averaged, and SAM variability for the past 500 y together with the 25 (26) negative
recalibrated, and revalidated directly against instrumental data. The aver- (positive) coupled ENSO/SAM events are also presented in Datasets S2 and
age correlation between ensemble members at each grid point was then S3. Additional instrumental climate data used in the paper are available in
calculated. Because opposite precipitation trends occur on both side of the the corresponding hosting websites.
highest Andes region (24° to 38°S), we produce two independent
15-member ensemble reconstructions on each sides of the Andes that were ACKNOWLEDGMENTS. We acknowledge the Servicio Meteorológico Nacio-
merged to create a final reconstruction. nal, Instituto Nacional de Tecnología Agropecuaria, Secretaría de Recursos
Hídricos, Autoridad Interjurisdiccional de Cuencas, Dirección Provincial del
Comparison between Historical and Tree Ring-Based Hydroclimate Reconstructions. Agua in Argentina, Administracion de Parques Nacionales Argentina, Estan-
cia Los Huemules, Dirección General de Aguas and Dirección Meteorológica
SEA, a nonparametric statistical technique, was used to determine the rela-
de Chile in Chile, and Servicio Nacional de Meterología e Hidrología in Boli-
tionships between the regional scPDSI reconstructions and drought/pluvial
via and Perú for providing quality-controlled daily and monthly climate data,
events from climate reconstructions based on historical records. The selected
which were essential for developing the reconstruction models. This study
scPDSI regions (red rectangles in Fig. 1A) for these analyses were the Al- was supported by the Agencia Nacional de Promoción Científica y Tecnológ-
tiplano (17° to 23°S; 66° to 70°W), central Chile (30° to 37°S; 70° to 72°W), ica, Argentina (PICT 2013-1880), Consejo Nacional de Investigaciones Científ-
and part of La Plata basin (31° to 37°S; 56° to 61°W). The historical sources icas y Tecnológicas (PIP 11220130100584) projects. The Inter-American
are precipitation records from Potosí, Bolivia (Altiplano; ref. 21), the snow Institute for Global Change Research CRN03, CRN2047 and CRN027 funded
and drought records from the Andes region of central Chile (ref. 46 and part of this study through grants (NSF Grant GEO-0452325). M.S.M., M.E.F.,
sources references therein), and the flood records in the city of Santa Fe and R.V. received partial support from Fondo Nacional de Desarrollo Cientí-
(mid-Paraná river, La Plata basin; ref. 29). In this analysis, the regional fico, Tecnológico y de Innovación Tecnológica, Perú (FONDECYT-BM-INC.INV
scPDSI records were used as the background time series and the dates of 039-2019); M.S.M. was supported in part by National Science Foundation
dry/wet years in the historical series as event years (Dataset S1). For each Grant AGS-1702789. M.S.M., R.V., A.M.S., I.A.M., D.A.C., A.L., R.U.-J., and
event, a 9-y lag window was used with the event year as the central value C.L. were supported in part by the THEMES project funded by the BNP
plus 4 y before and after the event. The 9-y scPDSI data were averaged for Paribas Foundation in the frame of its “Climate Initiative” program. J.E.S.,
each event to produce a mean pattern related to the historical event. The E.R.C., and A.P.W. were supported in part by NSF Grant AGS-1602581;
mean scPDSI pattern for the selected years was statistically evaluated for M.S.M., R.V., M.E.F., J.E.S., and A.P.W. were supported in part by NSF Grant
significance (95% confidence interval) by performing 1,000 Monte Carlo OISE-1743738; J.E.S. was supported in part by NSF Grant AGS-1805490. LDEO
contribution #8418. V.M. was supported in part by the Russian State Assign-
simulations (47) using random years from the scPDSI record.
ment Project 0148-2019-0004. D.A.C., J.B., C.L., A.G.-R., A.A.M., M.R.-B., C.A.,
G.V., D.A., and I.A.-B. were supported by the Chilean Research Council (FON-
Analysis of Hydrological Extreme Events. Dry/wet events were characterized by DECYT 1161381, 1201411, 1181956, and 11161061). D.A.C., A.L., A.G.-R.,
their intensity and spatial extent. We calculated the average of scPDSI for the R.U.-J., and A.A.M. were supported by the National Agency for Research
entire SADA domain from CE 1400 to 2000. Extreme intensity dry/pluvial and Development Chile (ANID/FONDAP/15110009). A.G.-R. was supported
events were determined by including scPDSI values lower/higher than the by the National Agency for Research and Development Chile (ANID/PAI/
95th and 5th percentiles, respectively. To determine large spatially wide- 77190101). M.E.F. was supported in part by ANPCyT PICT 2014-2797. R.U.-J.
spread dry/wet events, we first calculated the interannual fluctuations in the was partially supported by PAI/ANID/7818I20003. A.L., J.B., and R.U.-J. were
percent area of severe dry/wet conditions, i.e., the total number of grid points supported in part by the Fondecyt Grant 1171496.
1. M. R. Prieto et al., “Sequías extremas en Mendoza durante el siglo XX y principios del 3. R. D. Garreaud et al., The 2010–2015 megadrought in central Chile: Impacts on re-
XXI. Administración de la carencia y conflictos socio-políticos” in Medio Ambiente y gional hydroclimate and vegetation. Hydrol. Earth Syst. Sci. 21, 6307–6327 (2017).
Transformación Rural en la Argentina Contemporánea, G. Zarrilli, M. Ruffini, Eds. 4. R. Neukom et al., Facing unprecedented drying of the Central Andes? Precipitation
(Universidad Nacional de Quilmes, Buenos Aires, 2020), pp. 1–30. variability over the period AD 1000–2100. Environ. Res. Lett. 10, 84017 (2015).
Downloaded by guest on July 23, 2020
2. J. A. Arrese, “16th Session of the Commission on Sustainable Development” (United 5. C. Vera et al., Toward a unified view of the American monsoon systems. J. Clim. 19,
Nations, New York, 2008). 4977–5000 (2006).
EARTH, ATMOSPHERIC,
for climate reconstruction. Palaeogeogr. Palaeoclimatol. Palaeoecol. 281, 196–209
42. I. Harris, P. D. Jones, T. J. Osborn, D. H. Lister, Updated high‐resolution grids of
(2009).
monthly climatic observations—The CRU TS3.10 dataset. Int. J. Climatol. 34, 623–642
19. M. Vuille et al., A review of the South American monsoon history as recorded in stable
(2014).
isotopic proxies over the past two millennia. Clim. Past 8, 1309–1321 (2012).
43. K. Matsuura, National Center for Atmospheric Research Staff, The Climate Data
20. J. G. Palmer et al., Drought variability in the eastern Australia and New Zealand
Guide: Global (land) Precipitation and Temperature: Willmott & Matsuura, University
summer drought atlas (ANZDA, CE 1500–2012) modulated by the Interdecadal Pacific
Oscillation. Environ. Res. Lett. 10, 124002 (2015). of Delaware. https://climatedataguide.ucar.edu/climate-data/global-land-precipitation-
21. E. R. Cook, C. A. Woodhouse, C. M. Eakin, D. M. Meko, D. W. Stahle, Long-term aridity and-temperature-willmott-matsuura-university-delaware. Accessed 20 October 2017.
changes in the western United States. Science 306, 1015–1018 (2004). 44. National Center for Atmospheric Research Staff, The Climate Data Guide: GPCC:
22. E. R. Cook et al., Asian monsoon failure and megadrought during the last millennium. Global Precipitation Climatology Centre. https://climatedataguide.ucar.edu/climate-
Science 328, 486–489 (2010). data/gpcc-global-precipitation-climatology-centre. Accessed 20 September 2018.
23. E. R. Cook et al., Old World megadroughts and pluvials during the Common Era. Sci. 45. T. Melvin, K. Briffa, A “signal-free” approach to dendroclimatic standardization.
Adv. 1, e1500561 (2015). Dendrochronologia 26, 71–86 (2008).
24. J. Michaelsen, Cross-validation in statistical climate forecast models. J. Clim. Appl. 46. M. H. Masiokas et al., Snowpack variations since AD 1150 in the Andes of Chile and
Meteorol. 26, 1589–1600 (1987). Argentina (30°–37°S) inferred from rainfall, tree-ring and documentary records.
25. D. M. Meko, Dendroclimatic reconstruction with time varying subsets of tree indices. J. Geophys. Res. 117, D05112 (2012).
J. Clim. 10, 687–696 (1997). 47. C. Z. Mooney, R. D. Duval, Bootstrapping: A Nonparametric Approach to Statistical
26. A. Gioda, M. R. Prieto, Histoire des sécheresses andines: Potosi El Niño et le Petit Age Inference, (Sage University Paper Series on Quantitative Applications in the Social
Glaciaire. Meteorologie 8, 33–42 (1999). Sciences, Sage University, Newbury Park, CA, 1993).
27. B. Vicuña Mackenna, Ensayo Histórico Sobre el Clima de Chile: Desde los Tiempos 48. A. Cowling, P. Hall, M. J. Phillips, Bootstrap confidence regions for the intensity of a
Prehistóricos Hasta el Gran Temporal de Julio de 1877, (Imprenta del Mercurio, Val- Poisson point process. J. Am. Stat. Assoc. 91, 1516–1524 (1996).
paraíso, Chile, 1877). 49. M. Mudelsee, Climate Time Series Analysis: Classical Statistical and Bootstrap Meth-
28. R. Elizalde Mac-Clure, La Sobrevivencia de Chile. La Conservación de sus Recursos ods, (Atmospheric and Oceanographic Sciences Library, Springer, 2014), Vol. 51.
Naturales, (Minis. Agr., S.A.G., Santiago, Chile, 1970). 50. D. Wilks, Statistical Methods in the Atmospheric Sciences, (Academic, 2011).
29. M. R. Prieto, ENSO signals in South America: Rains and floods in the Parana River 51. M. S. Morales et al, South American Drought Atlas (SADA) 1400-2000 C.E. http://www.
during colonial times. Clim. Change 83, 39–54 (2007). cr2.cl/datos-dendro-sada/. Deposited 2 June 2020.
Downloaded by guest on July 23, 2020
Morales et al. PNAS | July 21, 2020 | vol. 117 | no. 29 | 16823