Content uploaded by Eduardo A. Agosta
Author content
All content in this area was uploaded by Eduardo A. Agosta on Jun 16, 2020
Content may be subject to copyright.
INTERNATIONAL JOURNAL OF CLIMATOLOGY
Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
Published online 21 February 2017 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/joc.5014
Climate change and precipitation variability over the western
‘Pampas’ in Argentina
Reinaldo A. Maenza,a,b,c*Eduardo A. Agostab,d,c and María L. Bettollia,c
aDepartamento de Ciencias de la Atmósfera y los Océanos, Universidad de Buenos Aires, Argentina
bEquipo Interdisciplinario para el Estudio de Procesos Atmosféricos en el Cambio Global [PEPACG], Facultad de Ingeniería y Ciencias
Agrarias, Ponticia Universidad Católica Argentina [UCA], Buenos Aires, Argentina
cConsejo Nacional de Investigaciones Cientícas y Tecnológicas [CONICET], Buenos Aires, Argentina
dFacultad de Ciencias Astronómicas y Geofísicas [FCAG], Universidad Nacional de La Plata [UNLP], Argentina
ABSTRACT: This work focuses on the analysis of spatial and temporal variability of precipitation in the central region
of southern Central Argentina (SCA), a climate transition area which has experienced an important agricultural expansion.
For this purpose, gauge station precipitation datasets available in the area were extensively used. The annual cycle shows
a dened dry season (May–August) and wet season (September –April). Wet season represents over 85% of annual totals.
A regionalization analysis of wet-season precipitation suggests ve subregions with spatially homogeneous precipitation
variability in SCA. Three out the ve subregions are located in central SCA. Conveniently devised precipitation indices for the
latter subregions show the presence of signicant precipitation jumps by the early 1970s, and to a minor extent, the mid-1960s.
Precipitation jumps are responsible for the observed long-term trends in central SCA, which explain positive precipitation
changes over 30– 40% of regional averages in the period 1922– 2012. The presence of stationary and non-stationary
components in SCA precipitation variability remotely connects the region mainly with variations in equatorial Pacic SSTs.
The assessment of greenhouse gases concentration effects on future projections of wet-season precipitation over central SCA
is investigated by means of multi-model analysis of historical experiment, and the representative concentration pathways 4.5
(RCP 4.5) and 8.5 (RCP 8.5), provided by the Coupled Model Intercomparison Project Phase 5. Results suggest an overall
increased precipitation, roughly 15% respect to present climate, under most severe future scenario.
KEY WORDS precipitation change; climate variability; the Pampas region; global warming; CMIP5; 1976–1977 climate shift
Received 4 August 2016; Revised 27 December 2016; Accepted 10 January 2017
1. Introduction
There are regions where the development of extensive agri-
culture and livestock industry has been strongly inuenced
by regional climate conditions. Advances in technology
allowed farming where climate conditions would be
unfavourable before. The Fifth Assessment Reports and
Special Reports (AR5) conrm that industrialization
and land cover changes have increased greenhouse gases
(GHG) concentrations particularly since mid-20th century,
inducing changes in regional climate conditions that may
affect regions with low infrastructure development (IPCC,
2013). The climate response to an invariably increas-
ing forcing such as the observed radiative imbalance is
expected to be steady trends and changes in climatic
variables such as sea level, precipitation, temperature, sea
ice, and others.
The current research focuses on time-space variability
in precipitation over southern Central Argentina (SCA)
* Correspondence to: R. A. Maenza, Ponticia Universidad Católica
Argentina, Ciencia y Técnica (UCA CyT), Alicia Moreau de Justo
1600, Suite 301, C1107AAZ, Ciudad de Buenos Aires, Buenos Aires,
Argentina. E-mail: pepacg@uca.edu.ar
region, located to the east of the Andes, roughly between
69∘–60∘W and 42∘–32∘S (Figure 1, solid line). It com-
prises La Pampa State, in central SCA, southern lands of
the states of San Luis and Córdoba, western Buenos Aires
State, northern Río Negro State, the eastern lands of the
states of Neuquén and Mendoza. Thus, it encompasses the
western portion of the ‘Pampas’ region in South America,
an Argentine traditional agricultural core region.
According to Köppen-Geiger’s world climate classica-
tion, updated by Kottek et al. (2006), the SCA is a region of
climate transition. To the southwest, the SCA is character-
ized by warm temperate climate with dry and warm sum-
mers (Csb). In the central lands of SCA, the climate shows
arid conditions with a steppe’s rainfall regime and cold
temperatures (BSk), while the northeast of SCA presents a
warm temperate climate with fully humid rainfall regime
and hot summers (Cfa). Thus, central SCA corresponds to
a transition zone between the dry ‘Pampas’ and the wet
‘Pampas,’ characterized by an intense precipitation gradi-
ent from the southwest to northeast.
The temporal variability of precipitation in SCA has
been subject to intensive research as being part of sub-
tropical Argentina, a vast region to the north of 40∘Sand
east of the Andean ranges (Castañeda and Barros, 1994;
© 2017 Royal Meteorological Society
446 R. A. MAENZA et al.
Figure 1. Administrative map of Argentina with the region of study:
Southern Central Argentina (SCA, solid line), and surrounding regions:
Central West Argentina (CWA, dash-dotted line), and Subtropical East-
ern Argentina (SEA, dashed line). Argentina’s state names in capital
letters: Buenos Aires (BA), Córdoba (CO), La Pampa (LP), Mendoza
(ME), Neuquén (NE), Río Negro (RN), San Luis (SL) and Santa Fe (SF).
Bottom right inset: map of South America and Argentina in black box.
Penalba and Vargas, 2001, 2004). Overall, the precipi-
tation time series in subtropical Argentina have shown
signicant trends and jumps towards wetter conditions
since at least mid-20th century. The preliminary results of
Minetti and Vargas (1998) showed the presence of trend
and jumps in total annual precipitation time series related
to stations located south of 15∘S and east of the Andes, in
the period 1900–1990. Despite the under-representation
of SCA because of the use of few stations within cen-
tral and western SCA, their research showed a jump to
more humid climate between 1950s and 1960s in the sta-
tions situated in eastern La Pampa State and northwestern
Buenos Aires State. Accordingly, Castañeda and Barros
(1994) and Barros et al. (2000) found positive annual rain-
fall trend in many localities in subtropical Argentina dur-
ing the period 1916–1990. The authors showed that sev-
eral sites underwent a total annual precipitation increase of
about 30% between 1956 and 1991. Thus, the total annual
precipitation rise amplied the availability of productive
lands (Manuel-Navarrete et al., 2009) over 100 000 km2,
favouring an important agricultural expansion over semi-
arid steppes (Barros et al., 2008).
Haylock et al. (2006) examined extreme daily rainfall
and annual rainfall rate indices in 1960–2005, and sug-
gested an increase in the annual total precipitation of
wet days and daily precipitation intensity over subtropical
Argentina, including a part of northern SCA represented in
their work by a small number of stations. Moreover, Rivera
et al. (2012) have found a decrease in the occurrence of
dry days in the same period, evidenced by signicant neg-
ative trends of 2 to 6 dry days per decade in northern SCA,
among other regions in central and southern Argentina.
Penalba and Vargas (2004) examined the total monthly
and annual precipitation from station data in the histori-
cal period 1901–2000 over subtropical eastern Argentina
(SEA, north of 40∘Sandeastof67
∘W, Figure 1, dashed
line). The south-west of SEA corresponds with the north-
east of SCA. Their results showed the presence of domi-
nant periodicities at interannual, decadal and interdecadal
bands in the vicinity of north-eastern SCA, suggesting
potential connection with El Niño-southern oscillation
(ENSO) phenomenon.
To the north-west, the SCA borders subtropical arid
plains immediately lying east of the subtropical Andes
in central-west Argentina (CWA, dash-dotted line in
Figure 1). The region underwent a sudden rise in summer
precipitation of about 25% by mid-1970s owing to the
inuence of the widely documented 1976/1977 climate
transition impact in the Americas (Ebbesmeyer et al.,
1991; Mantua et al., 1997; Agosta et al., 1999; Agosta
and Compagnucci, 2012). The climate shift was associated
with tropospheric circulation changes over southern South
America by perturbing the western ank of the subtropical
South Atlantic anticyclone at synoptic scale, enhancing
the moisture transport from tropical to subtropical lati-
tudes and, thus, causing the overall precipitation rise in
CWA (Compagnucci et al., 2002; Agosta and Compag-
nucci, 2008). More recently, Saurral et al. (2016) studied
the low-frequency variability and trends in centennial
precipitation time series in southern South America. The
authors found that the annual increase in precipitation
over SEA was mostly explained by summer variability.
Global Climate Models (GCMs) are commonly used
to examine and evaluate future regional climate changes
under scenarios of increasing GHGs. The Coupled Model
Intercomparison Project phase 5 (CMIP5) produce a
state-of-the-art multi-model dataset designed to improve
our knowledge of climate variability and climate change
(Taylor et al., 2012). Vera and Díaz (2014) analyzed the
anthropogenic effect on summer (December to February)
precipitation during the 20th century over southern South
America, by comparison of multi-ensemble mean trends
related to simulations from different CMIP5’s experi-
ments. They concluded that the anthropogenic forcing
holds a partial contribution in explaining the signicant
positive trends observed in southeastern South America
and negative trends observed in southern Andes.
Sillmann et al. (2013), using future climate simula-
tions under different emission scenarios from CMIP3 and
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 447
CMIP5, have shown that western/eastern SCA seems to
be characterized by a decrease/increase in the amount
of days with more than 1 mm of precipitation, accompa-
nied by an increment of the amount of consecutive dry
days over the entire SCA, mostly over western SCA, par-
ticularly in ensembles for high emission scenarios. The
latter results are congruent with that found by Penalba
and Rivera (2013) who by means of future projections
obtained from CMIP5 simulations calculated a drought
indicator to analyze drought future projections over south-
ern South America. They concluded that the occurrence of
short-term and long-term droughts would be more frequent
by the end of the 21st century, with shorter durations and
greater severities over most southern South America.
It is noteworthy that precipitation trends seem to have
reversed after the early 2000s in central and western SCA
(Pérez et al., 2011; Russian et al., 2015) and CWA (Agosta
and Compagnucci, 2012). The severe drought that hap-
pened from mid-2008 until mid-2009 had a strong impact
on crop production of Argentina’s agricultural area (Earth
Observatory, 2009; WMO, 2009). The latter evidences
the vulnerability of SCA to high-frequency uctuations in
rainfall. The vulnerability of farming activity to present cli-
mate, and to possible future climate changes, is evident,
suggesting the necessity to design adaptation measures.
Therefore, the current research has a twofold main aim:
rst, we are looking at the necessity to examine the
space-time variations of precipitation within SCA, using a
dense network of precipitation data available in the region,
which has not completely been exploited yet. This way, it
will be possible to device precipitation indices which ade-
quately describe the strong precipitation variability within
transitional climate of central SCA. Secondly, we will eval-
uate multi-model ensemble means for precipitation from
the RCP 4.5 and 8.5 simulations of the CMIP5 experi-
ments. The evaluation will be performed through historical
running showing good skill with observational precipita-
tion indices conveniently designed for central SCA. Thus,
we will be able to draw conclusions about future behavior
of precipitation.
The work is structured as follows: Section 2 presents the
observational network data and methodologies used for the
statistical analysis of the present-climate spatial-temporal
variability of precipitation, as well as the CMIP5 model
experiments selected to evaluate future conditions in pre-
cipitation in central SCA. Section 3 describes results
related to the spatial regionalization of precipitation and its
low-frequency temporal variability, and it discusses future
projections for precipitation. A summary and discussion
are offered in section 4.
2. Methodology and data
2.1. Data
2.1.1. Station precipitation data
We used monthly precipitation time series from 92 sta-
tions located within La Pampa State in the core of SCA
(see Figure S1, Supporting Information), provided by
the La Pampa State (Provincial) Water Administration
(PWA, available at http://www.apa.lapampa.gov.ar), over
the instrumental period available 1921– 2013. In addi-
tion, 45 stations over the SCA were used (see Figure
S1), provided by the Argentine National Weather Ser-
vice (NWS). Further information on the gauge stations are
summarized in Table S1. A quality control was applied
to both datasets in relation with possible outlier detec-
tion, using an inter-quartile criterion, and spatial inhomo-
geneities (Maenza, 2016).
2.1.2. GCM precipitation data
Monthly precipitation data from 15 models simulations
of the WCRP-CMIP5 were analyzed in relation with
three long-term experiments (Table 1). One is the his-
torical experiment, analyzed in the period 1922–2005,
which corresponds to the observational period available
in the region. The other two experiments are related to
future projections forced by specied concentrations in the
period 2006–2100, consistent with high emissions sce-
nario (RCP8.5) and midrange mitigation emissions sce-
nario (RCP4.5). The projections are both adopted by the
IPCC’s AR5. More details about experiments are avail-
able in Taylor et al. (2012). We also re-gridded all GCMs
outputs to a higher (0.5∘latitude and longitude) common
resolution grid, using bilinear interpolation (Accadia et al.,
2003). Models’ grid points were assigned to one subregion
following this criterion: a grid point is considered repre-
sentative of a subregion when its distance to a gauge station
compounding a subregion is lower than 0.5∘. Thus, every
grid point belongs to only one subregion.
2.2. Methods
2.2.1. Annual precipitation phases and regionalization
An unequal variance t-test for two-sample means (Ruxton,
2006) was applied to seasonal precipitation time series of
each single station, to determine grouping of months which
best discriminate between the wet phase and the dry phase
of the annual cycle. The analysis suggested two groupings
as most suitable: a dry season from May to August, and
a wet season from September to April. Note that a wet
season spans over 2 years. The reference year for a wet
season will be that one corresponding to the end of season.
To determine spatially homogeneous precipitation
subregions in each season, temporal variability of accu-
mulated precipitation time series was analyzed. The
rotated S-mode Principal Component Analysis (Richman,
1986; RSPCA, hereafter) was applied. The Varimax cri-
terion (Kaiser, 1958) was selected for rotation, ensuring
the orthogonality of the subregions and the correlation
matrix was used as similarity matrix for the input to
RSPCA. To guarantee extended spatial and temporal
homogeneity, the regionalization analysis was performed
in the period 1968–2011. The selection of stations was
done considering those stations with substantial quality of
data, due length of records and less than 5% of missing
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
448 R. A. MAENZA et al.
Table 1. CMIP5 models used in the current study.
Model name Institutes Resolution (lat. ×lon.) Total record
ACCESS1-0 Commonwealth Scientic and Industrial Research
Organization (Australia)
1.25∘×1.875∘1/1850–12/2100
ACCESS1-3 Commonwealth Scientic and Industrial Research
Organization (Australia)
1.25∘×1.875∘1/1850–12/2100
BCC-CSM1-1-m Beijing Climate Center, China Meteorological
Administration (China)
∼1.12∘×1.125∘1/1850–12/2100
CCSM4 National Center for Atmospheric Research (USA) ∼0.9∘×1.25∘1/1850–12/2100
CESM1-BGC National Center for Atmospheric Research (USA) ∼0.9∘×1.25∘1/1850–12/2100
CESM1-CAM5 National Center for Atmospheric Research (USA) ∼0.9∘×1.25∘1/1850–12/2100
CNRM-CM5 Centre National de Recherches Meteorologiques (France) ∼1.4∘×∼1.4∘1/1850– 12/2100
CSIRO-Mk3-6-0 Commonwealth Scientic and Industrial Research
Organization (Australia)
∼1.87∘×1.875∘1/1850–12/2100
HadGEM2-CC Met Ofce Hadley Center (UK) 1.25∘×1.875∘12/1934–11/2099
HadGEM2-ES Met Ofce Hadley Center (UK) 1.25∘×1.875∘12/1934–11/2124
MIROC5 Atmosphere and Ocean Research Institute, University of
Tokyo (Japan)
1.41∘×1.41∘1/1850–12/2100
MPI-ESM-LR Max Planck Institute for Meteorology (Germany) ∼1.87∘×1.875∘1/1850–12/2100
MPI-ESM-MR Max Planck Institute for Meteorology (Germany) ∼1.87∘×1.875∘1/1850–12/2100
MRI-CGCM3 Meteorological Research Institute (Japan) ∼1.12∘×1.125∘1/1850–12/2100
NorESM1-M Norwegian Climate Centre (Norway) ∼1.895∘×2.5∘1/1850–12/2100
Historical, RCP 4.5 and RCP 8.5 experiments were analyzed. Total record denote the period spanned by historical and RCPs experiments together.
data. Besides, based on the distance between stations, 34
of 137 gauge time series were selected for the analysis
to reduce redundant information in space (Figure 2).
Thus, the obtained spatial patterns (PC-loadings) are the
constituent subregions with associated temporal time
series (PC-scores). PC-loadings (correlations) exceeding
0.7 dene the boundary of a subregion. Every station
time series included within a subregion determined by
a PC-loading pattern for each season, were spatially
averaged to obtain a seasonal precipitation time series
representative of that subregion. To include information
from other stations and in a longer period, beyond the
regionalization analysis, correlation coefcient among
station time series and representative subregional time
series were estimated. Those stations showing correlation
over 0.7 were further included within each subregion.
2.2.2. Seasonal precipitation indices
Seasonal precipitation indices employed in this work
were formerly devised by Agosta et al. (1999) and
used by Compagnucci et al. (2002). The index was
computed as follows: at each station, the seasonal pre-
cipitation was expressed as the percentage deviation
from time mean in the baseline 1975–2004. Then, the
spatial average of every percentage deviation across
all the stations conforming a subregion determined the
seasonal precipitation index (SPI). Positive (negative)
values of SPI denote seasonal precipitation above (below)
regional-average. Hence, the index time series captures the
interannual-to-multidecadal variability of seasonal precip-
itation in a subregion (Agosta and Compagnucci, 2012).
A low-pass recursive 9-term Gaussian lter function
with Hamming window (Canavos, 1988, Mitchell et al.,
1966) was used to smooth seasonal climate indices and
to retain interdecadal variations. Throughout the analysis,
Figure 2. Spatial location of gauge stations used for the regionalization
analysis of accumulated precipitation in the wet season. Numbers indi-
cate the station name displayed in the Table S1.
SPI values were estimated from both observational and
modelled data.
The relationships and their stationarity among SPI
time series and climatic indices were estimated using a
20-year-window running correlation analysis (correlations
computed on a window of 20 years, moving throughout
the length of records). We used the canonical indices
related to El Niño phenomenon, namely, Nino1.2, Nino3,
Nino3.4 and Nino4 (Takahashi et al., 2011), as well as an
index for the Pacic Decadal Oscillation (PDO, Mantua
et al., 1997). We also considered an index for the tropical
Southern Atlantic variability (TSA) devised by Eneld
et al. (1999) but using the Extended Reconstructed Sea
Surface Temperature dataset (Huang et al., 2014, 2015;
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 449
Liu et al., 2014), and another index for the leading high
latitude variability mode known as the Southern Annular
Mode (SAM, also referred to as Antarctic Oscillation.
Gong and Wang, 1999, Marshall, 2003). All climatic
indices, excepting TSA, are available at http://www.esrl
.noaa.gov/psd/gcos_wgsp/Timeseries/. The statistical
signicance of the correlation was computed using a
Student’s t-distribution for a z-transformation of the
correlation (Wilks, 2006).
2.2.3. Jump, trend and periodicity analysis
The presences of jumps in the time series were carried
out by applying the Yamamoto’s statistics Jp(Yamamoto
et al., 1986, 1987). Further information about the latter
statistics can be found in Appendix S1. Trends were esti-
mated by means of the Linear least square (Wilks, 2006)
and Kendall’s Tau (Mann, 1945) methodologies in those
sub-periods established by Yamamoto’s statistics.
A spectral analysis of SPI time series corresponding
to each subregion was carried out by applying the con-
tinuous wavelet transform (Torrence and Compo, 1998;
CWT hereafter). The wavelet power spectrum gives infor-
mation of both the stationary and non-stationary com-
ponents of climatic index variability in time-frequency
domain. A Morlet wavelet function was used as mother
function as suggested by Grinsted et al. (2004) for geo-
physical purposes. Since time series with frequency dis-
tribution far from normal produces rather unreliable and
less signicant results in the wavelet power spectrum
(Grinsted et al., 2004), every climatic seasonal index was
tested for normality using a chi-square goodness-of-t test
(Wilks, 2006). Prior to applying CWT to standardized time
series, jumps and trends were ltered out. The tests and
statistics cited above were tested with a 95% condence
level.
2.2.4. GCMs validation
The assessment of model skills was carried out by com-
paring simulated precipitation outputs from the GCM his-
torical experiments to observational data. The compari-
son regarded their capabilities to represent some climatic
features of observed precipitation, such as the monthly
mean annual cycle, dispersion and extreme values of sea-
sonal accumulated precipitation totals, as well as their
low-frequency variability. GCMs show some deciencies
in representing amplitudes of precipitation over south-
ern South America (Bettolli and Penalba, 2014; Vera and
Díaz, 2014). In this work, we aimed to assess a quali-
tative evaluation of models’ performance in representing
different aspects of precipitation in SCA, which presents
a strong precipitation gradient. The use of as much inde-
pendent GCM information as possible can aid to over-
come the problem. We did not disregard any model ‘a
priori’, although the issue of independency among mod-
els is tackled by weighting the models when it comes
to estimate the multi-model ensemble means in Section
3.3.1.. To avoid potential biases on precipitation outputs
due to dependency among the models, models that share
a common model ‘family tree’ (Knutti et al., 2013) were
weighted in the estimation of the multi-model ensemble
means. Thus, multi-model ensemble means were estimated
with only one realization from each institution, i.e. we take
the average of outputs from similar models provided by
the same institution. Similarly, un-weighted multi-model
ensemble means were carried out. Results are comparable
(not shown) to the weighted approach being qualitatively
similar, which added robustness to the weighting analysis
(Knutti et al., 2010).
3. Results
3.1. Mean elds
As described in Section 1, central SCA is a region of
climate transition between the ecoregions wet ‘Pampas’
to the northeast and east, and dry ‘Pampas’ to the west
and southwest. The spatial distribution of the gauge net-
work provided by the NWS is quite uneven within SCA,
evidencing a substantial gap in the west of central SCA
(Figure 3(a)). The addition of gauge stations provided
by PWA covers up most of northeastern territories in
central SCA (Figure 3(b)). The annual mean precipitation
estimated using only the NWS gauge network offers an
overall smoothed precipitation eld with a gradient from
southwest to northeast in central SCA (Figure 3(c)). Mean
precipitation maxima between 800 and 1000 mm a year
are observed towards eastern and north-eastern bound-
aries, a secondary maximum centered about 800 mm is
observed towards the southwesternmost area close to
the Andes. Both maxima areas are separated by mean
precipitation minima along a northwest-southeast ori-
ented diagonal which denes the intense precipitation
gradient in central SCA, whose intensity is roughly about
500 mm in 425 km on average (a rate of 1.17 mm km−1).
Compared with the NWS gauge network, the mean
annual precipitation eld estimated using the full network
(NWS +PWA, Figure 3(d)) offers an overall similar spa-
tial pattern, although central SCA precipitation gradient
appears stronger and the isohyets sharper there where
most stations are located. Therefore, the inclusion of
more information provides more spatially detailed pre-
cipitation features in central SCA. Figures 3(e) and (f)
show the mean seasonal precipitation elds estimated
from the NWP-only database and NWS +PWA database,
respectively, in the wet-season. Overall the precipitation
patterns are quite like the annual means, corresponding
to each database. The main difference is the magnitude,
which is reduced in about 10–20% for the seasonal elds
in comparison to annual elds. In dry season, instead,
both database overall offers similar spatial features: a
gradual gradient towards drier conditions from east to
west (Figure 3(g) and (h)). Most of the SCA lies below
the isohyets of 150 mm, as illustrated by the mean eld
of the dry season. Because the wet season represents
on average over 85% of total annual precipitation in the
region, we will only analyze time series in wet season
onwards.
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
450 R. A. MAENZA et al.
Figure 3. Spatial location of gauge stations (dots) provided by the National Weather Service (NWS), (a) and the Provincial Water Administration
(PWA, b). Precipitation mean elds in the period 1981 –2010 for: (c) Annual totals from NWS-only stations, (d) annual totals from NWS +PWA
stations; (e) wet season from NWS-only stations, (f) wet season from NWS +PWA stations, (g) dry season from NWS-only stations, (h) dry season
from NWS +PWA stations. Contours every 50mm in the interval [0, 200) mm and every 100 mm in the interval [200, 900) mm.
3.2. Subregions in SCA during wet season
The regionalization analysis of precipitation for wet
season using RSPCA yielded four subregions in SCA
(Figure 4). One subregion (I) is located in central-northern
SCA, which is characterized by steep orographic relieves.
Another subregion (II) is situated in southeastern SCA, in
which an important agricultural expansion was observed
in the last decades. A third subregion (III) is located
in central-south SCA, presenting semiarid conditions
over plains. The location for the last subregion (IV) is in
northeastern SCA, hosting the most productive plains in
SCA area. Note that stations located over northeastern La
Pampa State were not classied within anyone of these
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 451
Figure 4. Location of the stations (markers) within the ve subregions
(indicated by roman numbers and corresponding marker type) obtained
by regionalization analysis of precipitation for wet season. Markers are:
for subregion I, plus sing; subregion II, dot; subregion III, asterisk;
subregion IV, diamond; and subregion V, cross.
four subregions (crosses in Figure 4). Nonetheless, these
stations show a robust structure function with correlation
coefcients among them over 0.7. Thus, a fth stand-alone
subregion (V) can be dened in northeastern La Pampa
State, surrounded by the semiarid plains to the west and
the most productive plains to the east and north.
Because our interest is to examine in detail precipitation
variability within the transitional climatic zone of central
SCA, subregions II, III and V will be the focus of further
analysis. These subregions have not been yet studied in
detail in the literature. Note that subregions II and V are
particularly related to agricultural expansion.
3.3. Temporal variability of climate indices
We have analyzed the low-frequency precipitation vari-
ability during wet season in the subregions II, III and
V through the corresponding precipitation indices (SPI),
described in Section 2.2. Figure 5 shows the SPI time
series and their smoothed time series. Throughout records,
dry conditions are predominant until the early 1970s.
The long-term dry epoch is, however, interrupted by
shorter dry-to-normal condition oscillations. From the
1970s to the early 2000s, alternating dry-and-wet periods
are observed. For the latter decade all the three subregions
show predominantly dry conditions in wet season.
Within the rst epoch of prolonged dry conditions,
it is possible to distinguish distinctive sub-periods of
drier conditions, as evidenced by the smoothed time
series. Thus, for the subregions II and V (Figure 5(a)
and (c)), and to a certain extent for III (Figure 5(b)),
four dry spells can be identied approximately between
1927–1939, 1946 –1952, 1958–1963 (subperiod not dis-
cernible for subregion III), and 1965–1972. These dry
sub-periods are interrupted by shorter and relatively wet-
ter spells. Within the period of prolonged wet conditions,
approximately between 1973 and 2002, it is noteworthy
Figure 5. Precipitation indices for wet season (SPI, bars) and the corre-
sponding time series smoothed by means of a low-pass recursive 9-term
Gaussian lter function with Hamming window (solid line) in subregions
II (a, upper panel), III (b, middle panel) and V (c, lower panel). Precipi-
tation expressed as percentage (%) deviations from regional mean on the
baseline 1975– 2004.
the presence of short dry sub-periods in the late 1980s and
the mid-1990s, although less intense than those recorded
before 1973. Overall, a common dry period is evident from
2003 for the three subregions. Subregion III shows greater
amplitudes. The long-term oscillation of predominant dry
and wet conditions identied above for central SCA, over-
all agree with those found by Compagnucci et al. (2002)
for CWA. In the latter region, Agosta and Compagnucci
(2012) found that the driest 30-year long-term interval in
the century expands between 1928 and 1957, whereas the
wettest 30-year long-term interval expands between 1973
and 2002. It is worthy to note that the dry spells observed
between the late 1920s and the mid-1950s are known as
the ‘Pampas Dust Bowl’ in the literature (Viglizzo and
Frank, 2006). Such an extended drought was undoubtedly
the severest drought in the past 100 years (Compagnucci
et al., 2002).
3.3.1. Climatic jumps analysis
To objectively identify possible climate jumps contained in
the SPI time series, the Yamamoto test is applied. Figure 6
shows the SPI time series (vertical bars) and the Yamamoto
statistic (Jp) values exceeding 1.0 for different windows
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
452 R. A. MAENZA et al.
Figure 6. Precipitation index for the wet season (SPI, vertical bars) in percentage (%, left axis) and Yamamoto statistics (markers, right axis) in
subregions II (a, upper panel), III (b, middle panel) and V (c, lower panel). The marker type denotes the window length (in years) used to estimate
the statistics in the analysis (bottom right inset). Only signicant values (over unity) of the statistics are shown.
(markers) and probability p=95%. Recall that when the
statistics overpass unity, it indicates years in which sig-
nicant climatic jump is detected. There is a sequence of
several years that can potentially be regarded as jumps
between the 1950s and the early 1980s for subregions II
and V (Figure 6(a) and (c)). Objectively a year with maxi-
mum in a sequence is considered the reference year for the
jump. Climatic jumps in subregion II seems to occur ear-
lier than in III and V, because the largest ratios are observed
between 1964 and 1968. Overall, the Yamamoto test yields
maximum ratios in 1964 and 1973 for subregion II, and
1973 for subregions V and III. Thus subregion II, in south-
eastern central SCA, undergoes an earlier jump towards
wetter conditions in the mid-1960s. The latter is manifes-
tation of the occurrence of moderate-to-strong wet years in
the wet seasons of 1964, 1968 and 1969, which were dry
years for subregions III and V. For the three subregions,
a jump towards drier conditions is detected by the early
2000s using a window of 6 years.
3.3.2. Trend analysis
When we look at trends in the full-record periods, the three
subregions show positive and signicant trends (Table 2).
The subregions II and V show signicant positive trends
under both tests, whereas the subregion III, only under the
LR test. The latter could be due to the recent dry epoch,
which is more severe in subregion III (see Figure 3(a)),
overcoming the overall positive trend. The changes in
precipitation due to trend are as great as +36%, +33% and
+40% of the corresponding areal averages in subregions
II, III and V, respectively. These percentages represent
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 453
Table 2. Linear trend analysis of precipitation index for wet season (SPI) in subregions II, III and V over the full period (1922– 2012
for II and V and 1939–2012 for III) and over the three main subperiods determined by the climatic jumps after Yamamoto’s test:
1922–1972 (1939 –1972 for subregion III), 1973–2002 and 2003– 2012.
Subperiod SPI 1922–1972 1973–2002 2003–2012 1922–2012
II −0.11/−4.62% +0.41/+22.23% −2.61/−26.1% +0.39/+35.49%
III +0.85/+28.9%
(1939–1972)
−0.47/−14.1% −4.1/−41.0% (LR) +0.45/+33.3%
(LR) (1939–2012)
V−0.01/−0.51% +0.59/+17.7% −0.39/−3.9% +0.44/+40.04%
Linear regression slope (in %/year) is indicated to the left of the slash and its associated change in precipitation over the subperiod is indicated to the
right of the slash. Values in bold indicate signicant trend according to the Linear Regression (LR)’s test and/or the Mann– Kendall (M– K)’s test,
both at 95% level of condence. If only signicant by one of the tests, the one is indicated in parenthesis. A change of 1% denotes an anomaly of
6.2, 3.8 and 7.6 mm with respect to the baseline mean in subregion II, III and V, respectively.
seasonal changes about 220, 127 and 304 mm for each
subregion, respectively, along their records.
Overall, trends within the sub-periods determined by
the climatic jumps appear not to be signicant. Only the
subregion III evidences signicant trend under both tests
along the long-term dry epoch before the rst climatic
jumps of the early 1970s. Note, however, that records in
the latter subregion start after 1938; therefore, its SPI time
series is missing information corresponding to the less dry
conditions of the early 1930s, as evidenced by the SPI time
series of subregions II and V. Thus, the strongly negative
values in the early records of the SPI time series for
subregion III are inuencing the slope in such a sub-period.
Again, this subregion shows a signicant and negative
trend (LR test) in early 2000s, giving a relevant negative
change in precipitation of −41%. Subregions II and V
also show negative trends in the sub-period, although not
signicant.
It is noteworthy that the long-term wet sub-period com-
mon to the three subregions, initiated by the climate jump
in the early 1970s, yields overall no signicant trends.
Only subregion II shows a signicant trend (M-K test)
when trend is estimated from the early 1960s, in the
sub-period 1964–2002 (results not shown). This trend can
be attributed to the presence of two jumps observed in
the sub-period, in 1964 and 1973. Note further that, after
removing the jumps, by subtracting the mean values cor-
responding to the periods dened by the climate jumps,
no signicant trends are obtained (Figure 7). Therefore,
we can state that the positive trends observed in central
SCA, which have given rise to enhanced precipitation in
about 30–40%, are substantially caused by sudden cli-
matic jumps observed by the 1970s, rather than by progres-
sive variability components. The following wavelet analy-
sis is thus performed on the ltered SPI time series.
3.3.3. Quasi-cycles in precipitation indices
The stationary and non-stationary components of
quasi-oscillations present in the ltered SPI time series
are further examined using CWT for subregions II, III and
V. Their wavelet power spectra are shown, respectively, in
the upper, middle and lower panels of Figure 8. Overall,
the wavelet analysis shows predominant non-stationary
components in the three subregions. Quasi-oscillations
Figure 7. As Figure 5, but for the SPI time series to which detected jumps
are ltered out.
common to the subregions can be observed in the peri-
odicity bands from 6 to 8 year between 1980 and 2000.
Periodicities from 2 to 5 year are observed in subre-
gions II and V between 1925 and 1945. Wavelet spectra
also show signicant quasi-cycles in the bandwidths of
4–6 year in the subperiods 1935–1945 and 1962 –1977
for subregion II (Figure 8(a)). The latter signal is also
present in subregion III in the subperiod 1987–1994
(Figure 8(b)) and in subregion V in the subperiod
1935–1945 (Figure 8(c)), roughly like in subregion
II. At decadal scales, quasi-cycles in the bandwidth
10–12 year are prominent in precipitation over subregions
II and V in the period 1940s and 1960s. For subregion V,
quasi-cycles in periodicities ranging from 16 to 18 year
appear from the early 1970s to the 2000s.
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
454 R. A. MAENZA et al.
Figure 8. Local wavelet power spectra of the SPI time series for the subregions II (a, upper panel), III (b, middle panel) and V (c, lower panel), using
the Morlet wavelet. Charts on top: standardized SPI time series for each subregion (solid line). Charts on the left side: Global wavelet spectrum for
each subregion (solid line), and condence interval (dashed line). The left axis is the Fourier period (year). Charts on the right side: The bottom axis
is time (year). Solid thick contour encloses 95% of condence for a red-noise process with a lag-1 coefcient obtained from each time series. The
cone of inuence (COI) where edge effects become important is shown as a lighter shade.
Some of the periodicities could be related to different
atmospheric-oceanic forcings, as they share quasi-cycles
in similar periodicity bands. The potential link between a
climatic forcing and precipitation in central SCA will be
tackled in the subsequent section.
3.3.4. Potential forcings
To look at related forcings, a 20-year-window running cor-
relation among the SPI time series and different climatic
indices at interannual scale are estimated (Figure 9). The
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 455
Figure 9. A 20-year-window running correlation analysis between detrended wet seasonal precipitation indices (SPI) in subregion II, III and V, and
detrended El Niño indices in different regions of the equatorial Pacic, represented by Nino1.2, Nino3, Nino3.4 and Nino4; the detrended Tropical
South Atlantic index (TSA) and the detrended Southern Annular Mode (SAM). Small dots indicate non-signicant correlation values, medium and
big dots indicate signicant values at 90% and 95% condence, respectively. The correlation coefcient (r) calculated for every full records are
shown in the right bottom corner of each panel together with the probability of signicance (p).
reference year of running correlation is centered in the
10th year. In addition, dispersion diagrams between SPI
time series and forcing indices will highlight the linear
relationship and their changes in those 20-year subperi-
ods, determined by the running correlation (Figure 10).
The analysis will give us an idea of the interannual
link between a forcing and precipitation, which is the
basis for any other relationship at longer scales. It
will further provide information of non-stationarity in
relationships.
Quasi-cycles in the biennial and intradecadal periodic-
ity bands could be related to the high (2–4 year) and low
(5-7 year) frequency components of the ENSO variability
(Penland et al., 2010). Subregions II and V show low cor-
relation with ENSO indices considering the full records
(all full-record correlation values are lower than 0.30),
although signicant maxima of running positive correla-
tion with the indices Nino1.2, Nino3 and PDO (the lat-
ter, not shown) appear by mid-1930s and the early 1970s
(r∼+0.5). Subregion II also shows signicant correlation
with indices Nino3.4 and Nino4 in the same subperiods.
Despite subregion III has a shorter temporal coverage,
signicant linear correlation between SPI and Nino1.2,
Nino3, Nino3.4 and PDO (the latter not shown) are also
found for the reference years centered about the early
1970s. Scatter plot diagram further conrm the direct rela-
tionship established between the equatorial Pacic SSTs
and precipitation in these subperiods (Figure 10, upper
panels). Such signicant direct relationships for subre-
gions II and V occur during the long-term dry conditions
recorded from the mid-1920s to mid-1940s, and with the
transition from dry to wet long-term conditions, recorded
between the early 1960s and late 1970s. Despite the rela-
tionship is non-stationary, running correlation values are
overall positive for ENSO indices, which is indicative of
the presence of a direct forcing at interannual scale in cer-
tain periods.
Another potential forcing at both intradecadal and
decadal variabilities could be the tropical South Atlantic
(Venegas et al., 1997; Tourre et al., 1999; Yuan and
Yonekura, 2011). The full-record correlations between
TSA index and SPIs show quite low values for the three
subregions. The most outstanding feature in the running
correlations is a phase change in the relationship of sig-
nicant correlation before and after the 1980s, especially
in subregions II and V. A direct link between the TSA
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
456 R. A. MAENZA et al.
Figure 10. Scatter diagrams between detrended SPI time series (abscissa) and detrended climate indices (ordered to origin): Nino3.4, TSA and SAM.
Correlation of the linear t (r) on left corner of each chart.
and precipitation is observed between the 1920s and
the mid-1950s, and between the 1960s and the 1970s.
Somehow in concordance with ENSO’s inuence: during
the decades of long-term dry conditions and during the
transitional decades towards wetter conditions in the
1970s. On the contrary, an inverse link between TSA
and precipitation is observed roughly from the 1980s
to mid-2000s over all central SCA. This remarkable
change in the relationship is also evidenced by the cor-
responding scatter diagrams (Figure 10, middle-lower
panels). Thus, the link between tropical South Atlantic
and precipitation looks not to be so straightforward.
Further analysis is required to fully understand such a
potential connection, which is left to future works. Prob-
ably, there may be contributions from remote ENSO’s
inuence upon the Atlantic SSTs (Kayano et al., 2013;
Bombardi et al., 2014). The running correlation between
ENSO’s indices and the TSA index point to this possibility
(not shown).
At intradecadal scale, a possible atmospheric forc-
ing could be the leading high latitude variability mode
depicted by the SAM index (Marshall, 2003). This mode
is known to be dominant at intradecadal and decadal
time scales (Yuan and Yonekura, 2011). The running
correlation analysis between SPI time series and the SAM
index are shown in the bottom panels of Figure 9. The
most prominent feature of this relationship is the wander-
ing behavior between negative and positive correlations
throughout the reference years. Correlation between SPI
in subregion II and V show positive and signicant corre-
lation values from mid-1930s to mid-1950s (see the peaks
in reference years about the mid-1940s). Later, the sign of
the relationship changes to negative from the early-1950s
to early-1970s (this signal is weaker for the subregion
V). From the 1970s onwards, the relationship between
the SPI time series and the SAM index behaves differ-
ently among the subregions. Thus, positive signicant
correlations are observed by the mid-1980s for subregion
II and subregion III. The non-stationary relationship is
further manifested by the corresponding scatter diagrams
(Figure 10, bottom panels). After the 1990s, the relation-
ship with SAM appears not to be signicant with any of
the subregions. Because of the wandering behavior of the
relationship between the precipitation variability and the
high latitude variability mode, it is apparent that there
may be other mechanisms coupling them that remain to be
elucidated.
Finally, since subregion V presents quasi-cycles close
to the bidecadal band, it would be interesting to explore
in future works if precipitation in the subregion could
be inuenced by the moon nodal 18-year cycle. In this
sense, Agosta (2014) found a bidecadal oscillation in CWA
precipitation that is present all along over a 100 years of
records, and the author related it with the inuence of the
lunar nodal cycle on the mid-latitude lower tropospheric
circulation.
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 457
3.4. Present conditions and future projections simulated
by GCMs
3.4.1. Present conditions simulated by GCMs
We will now evaluate the skill of models to reproduce the
main features of precipitation during the wet season in rela-
tion with mean, dispersion and extreme values. The annual
cycles of observed and modeled precipitation are shown in
Figure 11. In the subregions II and V, the model ensem-
ble underestimates the precipitation values during the wet
season, while dry season precipitation is overestimated
(Figure 11(a) and (c)). The underestimation during the wet
season is larger than the overestimation during the dry sea-
son, which is probably due to the deciency of the mod-
els in representing small scale processes involved in the
generation of summer rain (Bettolli and Penalba, 2014).
Even though the inter-model dispersion is large in the sub-
regions, the shape of the annual cycle is well captured
individually by each GCM. Correlations between observed
and modeled annual cycles are signicant at 95% in all
cases except for the MPI-ESM-MR and MRI-CGCM3
models in subregion II and the MPI-ESM-MR model in
subregion V (not shown). In the subregion III, instead,
the ability of the GCMs to capture the annual cycle is
low (Figure 11b). The weighted multi-model ensemble
mean shows an annual cycle less pronounced than the
observed one with an important overestimation of precipi-
tation values during the dry season. This is due to the high
inter-model spread. Some models tend to well represent
the shape of the annual cycle but they overestimate the pre-
cipitation values throughout the year. Some other models,
instead, show similar precipitation values to the observed
ones, but they completely misrepresent the shape of the
annual cycle, displaying maximum values during the dry
season. The fact is also reected by correlation between
observed and modeled precipitation annual cycles, where
10 out of 15 models show non-signicant correlation (not
shown). The incapability of some models to represent the
shapes of annual cycle in subregion III could probably be
due to the nearness of the Andes range, region in which
annual cycle shows austral winter precipitation maxima
and austral summer precipitation minima.
To assess model skill in reproducing climatic features in
the wet season, we examine box-plot diagrams (median,
lower quartile, upper quartile, and the maximum and min-
imum of data) for the observed and modeled data of sea-
sonal accumulated precipitation in the three subregions
(Figure 12, left panels). It is evident the high inter-model
spread in representing precipitation amounts. Coincident
with the annual cycle results, most models tend to under-
estimate precipitation values during wet season in the sub-
regions II and V (panels a and c), while they overestimate it
in the subregion III (panel b). Some models can reproduce
accumulated precipitation distribution reasonably well as
it is the case of NorESM1-M, CESM1-BGC and CCSM4
(both from NCAR) in subregions II and V, although with
interquartile range in the rst two models lower than
the observed. The BCC-CSM1-1-m and MRI-CGCM3
Figure 11. Monthly mean annual cycle obtained from monthly mean
precipitation totals in the period 1963– 2004, from observed (dotted
black line), simulated (gray lines with markers) and 15-model ensemble
(solid black line) data, spatially averaged over the subregions II (a, upper
panel), III (b, middle panel) and V (c, lower panel). Models’ acronyms
according to Table 1.
models reproduce with accuracy the features of the sea-
sonal precipitation distribution in the subregion III. The
results highlight the need to develop downscaling tech-
niques to improve regional precipitation totals analysis
over SCA.
Despite most models cannot properly simulate precip-
itation intensity and annual cycle, if we now examine
box-plot diagrams computed from SPI values obtained
with data from each GCM (Figure 12, right panels), most
of them tend to fairly well represent the observed SPI
distributions over the three subregions. The result sug-
gests that devising SPI through the percentage deviation
from time mean of regional precipitation is a manner in
which model inaccuracies in precipitation amplitudes are
avoided. Exceptions are the MPI-ESM-MR, that overesti-
mates the dispersion of SPI in the three subregions, and
the CESM1-BGC and CESM1-CAM5 models, belonging
to NCAR, that tend to underestimate the SPI dispersion,
especially in the subregion V (Figure 12(f)).
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
458 R. A. MAENZA et al.
(a) (d)
(b) (e)
(c) (f)
Figure 12. Box-plot diagrams (median, lower quartile, upper quartile, plus whiskers of maximum and minimum values of data) for the observed (left
panels) and simulated (right panels) data of accumulated precipitation in wet season and the SPI index in subregions II (panels a, d), III (panels b, e)
and V (panels c, f), respectively. Model names correspond to those described in Table 1. Data from the period 1963– 2004.
We further examine low-frequency SPI variations
through the 9-year smoothed SPI time series simulated by
every model in subregions II, III and V in the historical
period (Figure 13(a)–(c), respectively). The weighted
multi-model ensemble mean SPI time series are also
shown. A broad range of interdecadal variability is
discernible in the smoothed time series. The multi-model
ensemble means show an evident slow shift towards
increased precipitation along the 20th century in both
subregions. The percentage changes due to linear trend
simulated by the model ensemble means from unltered
data yield an increase in seasonal precipitation about 12%
for the three subregions over the period 1922– 2005, sig-
nicant at 95% of condence level. The simulated changes
estimate a third of the changes obtained from observations
(Section 3.2). The CCSM4 and CESM1-BGC are the
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 459
Figure 13. SPI time series smoothed by means of a low-pass recursive 9-term Gaussian lter function with Hamming window for subregions II,
III and V, obtained from the historical experiments (panels a, b, and c, respectively); for RCP4.5 experiment (panels d, e, and f, respectively); and
RCP8.5 experiment (panels g, h, and i, respectively) in gray lines with markers. Models’ acronyms according to Table 1, ensemble mean in solid
black line, and observed SPI smoothed time series in dotted black lines.
models that best reproduce the lower frequency variations
in seasonal precipitation in both subregions and they
behave similarly, probably because they belong to the
same model family. A spectral analysis conrms that
models generate temporal variations from interannual to
interdecadal scales, although with much weaker power
densities than observed (not shown). Thus overall, we nd
that models are able to capture low-frequency variations
in regional precipitation during the wet season.
The current assessment shows that simulated precipita-
tion annual cycles, amplitudes and climatic distributions
of accumulated totals in central SCA are little well rep-
resented by available GCMs. The use of a seasonal pre-
cipitation index expressed as percentage deviation from
time mean of regional precipitation (SPI) allows us to
avoid much of their inaccuracies. Lower frequency vari-
ations are well captured by most models along the obser-
vational period. Therefore, in the next section, a similar
multi-model analysis will be performed to evaluate future
projections using SPI.
3.4.2. Future projections
To compare future changes in seasonal precipitation
relative to present climate, projected SPI values are
also estimated as percentage deviation of accumulated
precipitation during wet season respect to the baseline
1975–2004, as used in previous sections. Future simula-
tions of low-frequency variations in SPI time series, from
RCP4.5 and RCP8.5 experiments from all the models for
subregions II, III and V, are shown in Figure 13(d)– (i),
in the period 2006–2099. The multidecadal variability is
remarkable in the three subregions for both experiments
throughout the 21st century. The multi-model ensemble
means make evident the continuity of a slow shift towards
positive values of seasonal precipitation, especially for
the RCP8.5 experiment, for which projected changes due
to linear trends along the 21st century yield an overall rise
about 15%, statistically signicant at a 95% condence
level.
Two subperiods of 30 years each are considered to eval-
uate changes in future climate, the middle (2040–2069)
and the late (2070–2099) 21st century. The distribution
of the weighted multi-model ensemble SPI means aver-
aged over the mid- and late 21st century is examined using
box-plot diagrams (Figure 14), corresponding to the RCP
4.5 (left panels) and RCP 8.5 (right panels) scenarios for
the subregions II, III and V (upper, middle and lower pan-
els, respectively).
The box-plots suggest increased precipitation in both
future periods and experiments for the three subregions
(Figure 14) with larger values for the RCP 8.5, which is
consistent with the changes due to trend (Figure 13). In the
late 21st century, the medians of the RCP 4.5 simulations
show precipitation rises about 8%, 4% and 6% respect to
present climate, in subregions II, III and V, respectively.
The medians of the RCP 8.5 simulations show rises about
10, 15 and 9% in subregions II, III and V, respectively. Dis-
persion of modeled means is reected by the quartiles and
extremes. As can be noticed in Figure 14, the rst and third
quartiles increase from one period to the other, for both
experiments in subregions II and V, but only the third quar-
tile for subregion III (in RCP8.5, 2070–2099). The quartile
increases are larger for RCP 8.5 simulations in the three
subregions. Note that for the subregion III, a considerable
increase in its inter-quartile range is observed for simu-
lations in late 21st century (panel e), which evidences a
greater inter-model spread. Boxplots diagrams show that
most models project positive mean SPI values, suggesting
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
460 R. A. MAENZA et al.
(a) (d)
(b) (e)
(c) (f)
Figure 14. Box-plot diagrams (median, lower quartile, upper quartile, plus whiskers of maximum and minimum values of data) of SPI’s mean
values corresponding to the 15 CMIP5 simulations for the RCP 4.5 (left panels) and 8.5 (right panels) experiments, for the periods 2040– 2069 and
2070– 2099, in the subregions II (a, d), III (b, e) and V (c, f).
a progressive rise in precipitation during wet season over
central SCA by the end of the 21st century.
Thus, future projections under the most severe scenario
indicate an overall increase in precipitation, roughly about
15%, for wet season over SCA along the 21st century.
Note that such an increase in precipitation is referred to the
present conditions (1975–2004), which are characterized
by mostly wet conditions in the region (see Section 3.3).
4. Summary and discussion
We have rstly aimed at studying spatial and temporal vari-
abilities of precipitation over SCA, a region of the western
‘Pampas’, spanning between 32∘–42∘S and 72∘–60∘W, in
the instrumental period 1922–2012. Similarly, secondly,
we have assessed the behavior of projected precipitation
using multi-model ensemble from CMIP5-GCM experi-
ments forced by specic GHGs concentration, consistent
with high emissions scenario (RCP8.5) and midrange mit-
igation emissions scenario (RCP4.5).
Thereby, a seasonality analysis shows that the wet phase
of annual cycle comprises the months from September to
April, and the dry phase, from May to August. Wet sea-
son represents over 85% of the annual total over SCA. In
wet season, a regionalization analysis yielded ve subre-
gions in which temporal precipitation variability can be
discriminated. We have further analyzed precipitation vari-
ability in subregions II, III and V which encompass cen-
tral SCA, characterized by being a territory of agricultural
expansion during the second half of the last century. We
found that precipitation variability in wet season shows
an overall long-term dry period between the 1920s and
1960s, followed by a long-term wet period between the
1970s and the early 2000s, and the initiation of a dry period
appears afterwards. Amplitudes are more pronounced in
subregion III.
In consequence, the long-term subregional trends gen-
erate signicant increases in precipitation along records
roughly about 33–40% in central SCA. Furthermore,
we have identied a signicant simultaneous hydro-
climate jump in all three subregions of central SCA
in 1973. Because of the jumps in precipitation time
series, the observed precipitation trends towards wetter
conditions throughout the past century appears not to
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 461
be a consequence of secular forcings, but of detected
jumps.
It is interesting to note that, in connection with central
SCA, the CWA region also underwent a signicant wet
seasonal precipitation jump in 1973 (Compagnucci et al.,
2002; Agosta and Compagnucci, 2008). Because of the
interdecadal variability present in CWA precipitation, the
authors attributed this 1973 jump to the 1976/1977 climate
shift observed in the central equatorial Pacic SSTs that
affected climate worldwide. The climate shift is coincident
with a phase change of the PDO index, which has deter-
mined a ‘La Niña-like’ decadal regime before and an ‘El
Niño-like’ decadal regime afterwards (Ebbesmeyer et al.,
1991; Mantua et al., 1997; Solomon et al., 2007). Further-
more, given the outstanding magnitude of the mid-1970s
shift, Jacques-Coper and Garreaud (2015) agreed with
Agosta and Compagnucci (2012) that the climate shift
effects on southern South America climates seem to be
unprecedented during the 20th century. In our analysis, we
have found that the equatorial Pacic basin also plays a
relevant role in regulating precipitation variability at inter-
annual scale during the long-term dry period observed
from the 1920s to the 1960s, and during the transitional
epoch related to the climatic shift in the mid-1970s. Hence,
the change towards wetter conditions in central SCA can
be regarded as another manifestation of the climate shift
in southern South America. Note, however, that the PDO
index also has previously changed its phase by the late
1940s, a change that is not recorded by precipitation vari-
ability in central SCA. Therefore, to a certain extent, we
cannot think of PDO variability as a direct driver alone of
the regional precipitation in the long term. Other possible
factors involved in the change could be non-linear internal
variability of the climate system, as well as enhanced GHG
forcing related to regional Hadley cell changes and/or solar
variability response (Solomon et al., 2007).
Model simulations further showed that while the domi-
nant inuence comes from the Pacic basin, the Atlantic
inuence can partially explain a large transition from dry
to wet decades over west and central Argentina during the
beginning of the 1970s (Barreiro et al., 2014). The integral
inuence of both oceanic basins enhances the moisture
transport and convergence in west and central Argentina
and, together with enhanced evaporation, increased the
rainfall after 1970 (Barreiro et al., 2014). Notwithstand-
ing, the particular role of the tropical South Atlantic upon
precipitation variability in SCA requires further analysis
to establish potential connections. In addition, potential
atmospheric teleconnections changes affecting precipita-
tion variability in central SCA relating with precipitation
jumps detected by the mid-1960s in subregion II, by
the 1970s and the early 2000s in subregions II, III and
V, will be subject to further analysis in future works.
To this respect, for instance, Agosta and Compagnucci
(2012) found that the precipitation variability over CWA
can be associated with barotropic quasi-stationary waves
emanating from the tropical Indian Ocean and the South
Pacic from the beginning of the 20th century until
mid-1970s. After the 1976/1977 climate shift, however, a
complete change in the teleconnection is found.
The assessment of GHGs concentration effects on future
projections of wet-season precipitation over central SCA
has been undertaken for the 21st century, as well as for
two 30-year periods: the middle (2040–2069) and late
(2070–2099) 21st century. Historical experiment was used
to evaluate the skills of 15 GCMs (Table 1) to reproduce
the present hydrological conditions. Simulated trends by
multi-model ensemble means yield an overall increase of
precipitation about 11% during wet season in the instru-
mental period, which is a third of the observational change
due to trends. Note that overall, GCMs are able to generate
some of the multi-decadal variability present in regional
precipitation. Future projections under the most severe sce-
nario indicate a linear increase in precipitation for wet sea-
son over SCA, roughly about 15%, emerging from the mul-
tidecadal variability by along the 21st century. The results
show that the future climate would be slightly wetter than
the present climate, which is one already wet. It is in such
a context of a regional climate wetter than the present con-
ditions in which Penalba and Rivera (2013) further found
that droughts would be more frequent by the end of the
21st century, with shorter durations and greater severities
over much of southern South America, including SCA. In
this sense, vulnerability associated with long-term uctua-
tions from wet to dry conditions, as the one observed after
the early 2000s, could be as severe as in the latter decade.
Acknowledgements
The funding for this research provided by projects,
PICT-2013 N 0043 from ANPCyT, PIP N∘112-201301-
00400 from CONICET, and UBACyT 2014– 2017
20020130200142BA from University of Buenos Aires.
Many thanks to the Carmelite Order.
Supporting information
The following supporting information is available as part
of the online article:
Figure S1. Location of the gauge stations.
Tabl e S 1. Information about gauge station.
Appendix S1. Yamamoto’s test.
References
Accadia C, Mariani S, Casaioli M, Lavagnini A, Speranza A. 2003. Sen-
sitivity of precipitation forecast skill scores to bilinear interpolation
and a simple nearest-neighbor average method on high-resolution ver-
ication grids. Weather Forecast. 18: 918–932.
Agosta EA. 2014. The 18.6-year nodal tidal cycle and the bi-decadal
precipitation oscillation over the plains to the east of subtropi-
cal Andes, South America. Int. J. Climatol. 34: 1606– 1614, doi:
10.1002/joc.3787.
Agosta EA, Compagnucci RH. 2008. Procesos Atmosféricos/Oceánicos
de baja frecuencia sobre la cuenca sudoeste del Atlántico Sur y la
variabilidad de la precipitación en el Centro-Oeste de Argentina.
Geoacta 33: 21– 32.
Agosta EA, Compagnucci RH. 2012. Central-west Argentina summer
precipitation variability and atmospheric teleconnections. J. Clim. 25:
1657– 1677.
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
462 R. A. MAENZA et al.
Agosta EA, Compagnucci RH, Vargas W. 1999. Cambios en el régi-
men interanual de la precipitación estival en la región Centro-Oeste
Argentina. Meteorológica 24(1): 63–84.
Barreiro M, Díaz N, Renom M. 2014. Role of the global oceans and
land– atmosphere interaction on summertime interdecadal variability
over northern Argentina. Clim. Dyn. 42: 1733– 1753.
Barros V, Castañeda ME, Doyle M. 2000. Recent precipitation trends in
Southern South America to the East of the Andes: an indication of a
mode of climatic variability. In Southern Hemisphere Paleo and Neo-
climates, Smolka P, Volkheimer W (eds). Springer: Berlin/Heidelberg,
Germany, 187–206.
Barros VR, Doyle ME, Camilloni IA. 2008. Precipitation trends in
southeastern South America: relationship with ENSO phases and with
low-level circulation. Theor. Appl. Climatol. 93: 19– 33.
Bettolli ML, Penalba OC. 2014. Synoptic sea level pressure
patterns– daily rainfall relationship over the Argentine Pampas
in a multi-model simulation. Meteorol. Appl. 21(2): 376– 383.
Bombardi RJ, Carvalho LM, Jones C, Reboita MS. 2014. Precipitation
over eastern South America and the South Atlantic Sea surface temper-
ature during neutral ENSO periods. Clim. Dyn. 42(5-6): 1553–1568,
doi: 10.1007/s00382-013-1832-7.
Canavos GC. 1988. Probabilidad y estadística: Aplicaciones y métodos.
McGraw Hill: Interamericana, México.
Castañeda M, Barros V. 1994. Las tendencias de la precipitación en el
cono sur de América al este de los Andes. Meteorologica 19: 23– 32.
Compagnucci RH, Agosta EA, Vargas MW. 2002. Climatic change
and quasi-oscillations in central-west Argentina summer precipitation:
main features and coherent behavior with southern African region.
Clim. Dyn. 18: 421– 435.
Earth Observatory. 2009. Drought in Argentina. Naturals Hazards.
http://earthobservatory.nasa.gov/NaturalHazards/view. php?id=37105
(accessed 1 March 2010).
Ebbesmeyer CC, Cayan DR, McLain DR, Nichols FH, Peterson DH,
Redmond T. 1991. 1976 step in the Pacic climate: Forty environ-
mental changes between 1968–1975 and 1977–1984. In Proceedings
of the Seventh Annual Pacic Climate (PACLIM) Workshop, Betan-
court JL, Tharp VL, (eds), Interagency Ecological Studies Program
Technical Report No. 26, California Department of Water Resources,
Sacramento, CA.
Eneld DB, Mestas AM, Mayer DA, Cid-Serrano L. 1999. How ubiqui-
tous is the dipole relationship in tropical Atlantic sea surface temper-
atures? J. Geophys. Res. 104: 7841– 7848.
Gong DY, Wang SW. 1999. Denition of antarctic oscillation index.
Geophys. Res. Lett. 26: 459– 462.
Grinsted A, Moore JC, Jevrejeva S. 2004. Application of the cross
wavelet transform and wavelet coherence to geophysical time series.
Nonlinear Processes Geophys. 11: 561– 566.
Haylock MR, Peterson TC, Alves LM, Ambrizzi T, Anunciação YMT,
Baez J, Barros VR, Berlato MA, Bidegain M, Coronel G, Corradi
V, Garcia VJ, Grimm AM, Karoly D, Marengo JA, Marino MB,
Moncunill DF, Nechet D, Quintana J, Rebello E, Rusticucci M,
Santos JL, Trebejo I, Vincent LA. 2006. Trends in total and extreme
South American rainfall in 1960– 2000 and links with sea surface
temperature. J. Clim. 19: 1490– 1512, doi: 10.1175/JCLI3695.1.
Huang B, Banzon VF, Freeman E, Lawrimore J, Liu W, Peterson TC,
Smith TM, Thorne PW, Woodruff SD, Zhang HM. 2014. Extended
reconstructed sea surface temperature version 4 (ERSST.v4): Part
I. Upgrades and intercomparisons. J. Clim. 28: 911– 930, doi:
10.1175/JCLI-D-14-00006.1.
Huang B, Thorne P, Smith T, Liu W, Lawrimore J, Banzon V, Zhang H,
Peterson T, Menne M. 2015. Further exploring and quantifying uncer-
tainties for extended reconstructed sea surface temperature (ERSST)
version4(v4).J. Clim. , doi: 10.1175/JCLI-D-15-0430.1.
IPCC. 2013. Climate Change 2013. The Physical Science Basis. Contri-
bution of Working Group I to the Fifth Assessment Report of the Inter-
governmental Panel on Climate Change. Summary for Policymakers,
Stocker TF, Qin D, Plattner GK, Tignor M, Allen SK, Boschung J,
Nauels A, Xia Y, Bex V, Midgley PM (eds). Cambridge University
Press: Cambridge, UK and New York, NY.
Jacques-Coper M, Garreaud RD. 2015. Characterization of the 1970s
climate shift in South America. Int. J. Climatol. 35: 2164– 2179.
Kaiser HF. 1958. The Varimax criterion for analytic rotation in factor
analysis. Psychometrika 23(3): 187– 200.
Kayano MT, Andreoli RV, Ferreira de Souza RA. 2013. Relations
between ENSO and the South Atlantic SST modes and their effects
on the South American rainfall. Int. J. Climatol. 33: 2008– 2023, doi:
10.1002/joc.3569.
Knutti R, Abramowitz G, Collins M, Eyring V, Gleckler PJ, Hewitson
B, Mearns L. 2010. Good practice guidance paper on assessing and
combining multi model climate projections. In Meeting Report of
the Intergovernmental Panel on Climate Change Expert Meeting on
Assessing and Combining Multi Model Climate Projections, Stocker
TF, Qin D, Plattner G-K, Tignor M, Midgley PM (eds). IPCC Working
Group I Technical Support Unit, University of Bern: Bern.
Knutti R, Masson D, Gettelman A. 2013. Climate model genealogy:
generation CMIP5 and how we got there. Geophys. Res. Lett. 40:
1194– 1199, doi: 10.1002/grl.50256.
Kottek M, Grieser J, Beck C, Rudolf B, Rubel F. 2006. World Map of
the Köppen-Geiger climate classication updated. Meteorol. Z. 15:
259– 263, doi: 10.1127/0941-2948/2006/0130.
Liu W, Huang B, Thorne PW, Banzon VF, Zhang HM, Freeman E,
Lawrimore J, Peterson TC, Smith TM, Woodruff SD. 2014. Extended
reconstructed sea surface temperature version 4 (ERSST.v4): Part II.
Parametric and structural uncertainty estimations. J. Clim. : 931– 951,
doi: 10.1175/JCLI-D-14-00007.1.
Maenza . 2016. Analysis of the spatial and temporal variability of
precipitation in southern Central Argentina and possible forcings of
the coupled atmosphere– ocean system at interanual-to-interdecadal
scale. PhD Dissertation, School of Sciences, State Buenos Aires
University, Argentina. http://digital.bl.fcen.uba.ar/.
Mann HB. 1945. Non-parametric tests against trend. Econometrica 13:
163– 171.
Mantua NJ, Hare SR, Zhang Y, Wallace JM, Francis RC. 1997. A Pacic
interdecadal climate oscillation with impacts on salmon production.
Bull. Am. Meteorol. Soc. 78: 1069– 1079.
Marshall GJ. 2003. Trends in the Southern Annular Mode from observa-
tions and reanalyses. J. Clim. 16: 4134– 4143.
Minetti JL, Vargas WM. 1998. Trends and jumps in the annual precip-
itation in South America, south of the 15∘S. Atmósfera (Mexico) 11:
205– 221.
Mitchell, JM, Dzerdzeevskii B, Flohn H, Hormeyr WL, Lamb HH, Rao
KN and Wallen CC. 1966. Climate change. WMO Technical Note No.
79, 33– 42, World Meteorological Organization, Geneva, Switzerland.
Penalba OC, Rivera JA. 2013. Future changes in drought charac-
teristics over Southern South America projected by a CMIP5
multi-model ensemble. Am. J. Clim. Change 2: 173– 182, doi:
10.4236/ajcc.2013.23017.
Penalba OC, Vargas WM. 2001. Propiedades de décits y excesos de
precipitación en zonas agropecuarias. Meteorologica 26: 39– 55.
Penalba OC, Vargas WM. 2004. Interdecadal and interannual varia-
tions of annual and extreme precipitation over central-northeastern
Argentina. Int. J. Climatol. 24: 1565– 1580.
Penland C, Sun DZ, Capotondi A, Vimont DJ. 2010. A brief introduc-
tiontotheNiñoandLaNiña.InClimate Dynamics: Why Does Climate
Vary ? Sun DZ, Bryan F (eds). American Geophysical Union: Wash-
ington, D.C., 216 pp, doi: 10.1029/2008GM000846.
Pérez S, Sierra E, López E, Nizzero G, Momo F, Massobrio M.
2011. Abrupt changes in rainfall in the Eastern area of La Pampa
Province, Argentina. Theor. Appl. Climatol. 103: 159–165, doi:
10.1007/s00704-010-0290-y.
Richman MB. 1986. Rotation of principal components. J. Climatol. 6:
293– 335.
Rivera JA, Penalba OC, Betolli ML. 2012. Inter-annual and inter-decadal
variability of dry days in Argentina. Int. J. Climatol. 33(4): 834–842.
Russian G, Agosta E, Compagnucci RH. 2015. Variaciones en baja
frecuencia de la precipitación estacional en la región Pampa amarilla
y posibles forzantes. Meteorologica 40(1): 17– 42.
Ruxton GD. 2006. The unequal variance t-test is an underused alternative
to Student’s t-test and the Mann– Whitney U test. Behav. Ecol. 17(4):
688– 690.
Saurral RI, Camilloni IA, Barros VR. 2016. Low-frequency variability
and trends in centennial precipitation stations in southern South Amer-
ica. Int. J. Climatol. , doi: 10.1002/joc.4810.
Sillmann J, Kharin VV, Zwiers FW, Zhang X, Bronaugh D. 2013.
Climate extremes indices in the CMIP5 multimodel ensem-
ble: part 2. Future climate projections. J. Geophys. Res. 118(6):
2473– 2493.
Solomon S, Qin D, Manning M, Marquis M, Averyt K, Tignor MMB,
Miller HL Jr, Chen Z (eds). 2007. Climate Change 2007: The Phys-
ical Science Basis. Cambridge University Press: Cambridge, UK,
996 pp.
Takahashi K, Montecinos A, Goubanova K, Dewitte B. 2011. ENSO
regimes: reinterpreting the canonical and Modoki El Niño. Geophys.
Res. Lett. 38: L10704, doi: 10.1029/2011GL047364.
Taylor KE, Stouffer RJ, Meehl GA. 2012. An overview of CMIP5 and
the experiment design. Bull. Am. Meteorol. Soc. 93: 485– 498, doi:
10.1175/BAMS-D-11-00094.1.
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
CHANGE AND PRECIPITATION VARIABILITY IN WESTERN ‘PAMPAS’ 463
Torrence C, CompoGP. 1998. A practical guide to wavelet analysis. Bull.
Am. Meteorol. Soc. 79: 61– 78.
Tourre YM, Rajagopalan B, Kushnir Y. 1999. Dominant patterns of
climate variability in the Atlantic Ocean during the last 136 years.
J. Clim. 12: 2285– 2299.
Venegas SA, Mysak LA, Straub DN. 1997. Atmosphere-ocean coupled
variability in the South Atlantic. J. Clim. 10: 2904– 2920.
Vera CS, Díaz L. 2014. Anthropogenic inuence on summer precipita-
tion trends over South America in CMIP5 models. Int. J. Climatol.
35(10): 3172– 3177, doi: 10.1002/joc.4153.
Viglizzo EF, Frank FC. 2006. Ecological interactions, feedbacks, thresh-
olds and collapses in the Argentine Pampas in response to climate and
farming during the last century. Quat. Int. 158: 122–126.
Wilks DS. 2006. Statistical Methods in the Atmospheric Sciences.Else-
vier: Oxford, UK, 628 pp.
World Meteorological Organization – WMO. 2009. The warmest
decade 2000– 2009. Press Release No. 869, 2009. http://www.wmo
.int/pages/mediacentre/press_releases/pr_869_en.html (accessed
March 2010).
Yamamoto R, Iwashima T, Sanga NK. 1986. An analysis of climatic
jump. J. Meterol. Soc. Jpn. 64(2): 273–280.
Yamamoto R, Iwashima T, Sanga NK. 1987. Detection of the Cli-
matic Jumps, Presented at XIX IUGG General Assembly, Vancouver,
Canada.
Yuan X, Yonekura E. 2011. Decadal variability in the Southern Hemi-
sphere. J. Geophys. Res. 116: D19115, doi: 10.1029/2011JD015673.
© 2017 Royal Meteorological Society Int. J. Climatol. 37 (Suppl.1): 445– 463 (2017)
