Climate Dynamics

https://doi.org/10.1007/s00382-019-04959-y

Dynamical downscaling over the complex terrain of southwest South

America: present climate conditions and added value analysis
Deniz Bozkurt1   · Maisa Rojas1,2 · Juan Pablo Boisier1 · Roberto Rondanelli1,2 · René Garreaud1,2 · Laura Gallardo1,2

Received: 30 December 2018 / Accepted: 26 August 2019

© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
This study evaluates hindcast simulations performed with a regional climate model (RCM, RegCM4) driven by reanalysis
data (ERA-Interim) over the Pacific coast and Andes Cordillera of extratropical South America. A nested domain configu-
ration at 0.44◦ ( ∼ 50 km) and 0.09◦ ( ∼ 10 km) spatial resolutions is used for the simulations. RegCM4 is also driven by a
global climate model (GCM, MPI-ESM-MR) on the same domain configuration to asses the added values for temperature
and precipitation (historical simulations). Overall, both 10 km hindcast and historical simulation results are promising and
exhibit a better representation of near-surface air temperature and precipitation variability compared to the 50 km simula-
tions. High-resolution simulations suppress an overestimation of precipitation over the Andes Cordillera of northern Chile
found with the 50 km simulations. The simulated daily temperature and precipitation extreme indices from 10 km hindcast
simulation show a closer estimation of the observed fields. A persistent warm bias ( ∼ + 4 ◦ C ) over the Atacama Desert in 10
km hindcast simulation reveals the complexity in representing land surface and radiative processes over the desert. Difficul-
ties in capturing the temperature trend in northern Chile are notable for both hindcast simulations. Both resolutions exhibit
added values for temperature and precipitation over large parts of Chile, in particular, the 10 km resolves the coastal-valley
Andes transitions over central Chile. Our results highlight that resolutions coarser than 50 km (e.g., GCMs and reanalysis)
miss important climate gradients imposed by complex topography. Given that the highest spatial resolution of the current
regional simulations over the South America is about 50 km, higher resolutions are important to improve our understanding
of the dynamical processes that determine climate over complex terrain and extreme environments.

Keywords  Model evaluation · Temporal-spatial scale analysis · Climate variability · Chile · Patagonia · Atacama Desert

1 Introduction due to their coarse spatial resolutions ( ∼ 100–200 km), a

number of initiatives to coordinate regional climate model
Regional-to-local scale climate information has become an (RCM) simulations have arisen in recent years to provide
issue of central importance, particularly to assess vulner- regional-to-local scale climate change information. Within
ability to climate change and to integrate adaptation needs at these initiatives, global multimodel programs such as the
regional and local levels. Given that global climate models Coordinated Regional Downscaling Experiment (COR-
(GCMs) generally fail to represent regional-scale features DEX) (Giorgi et al. 2009; Jones et al. 2011) are important
to further improve the quality of the RCM results but also to
explore the major shortcomings and strengths in the RCMs
Electronic supplementary material  The online version of this (Solman 2013).
article (https​://doi.org/10.1007/s0038​2-019-04959​-y) contains Three important initiatives (CORDEX, CLARIS-LPB,
supplementary material, which is available to authorized users. CREMA) led to an important progress in RCM applications
* Deniz Bozkurt over the South America for RCM evaluation studies as well
[email protected] as for long-term climate variability with an emphasis on
the Plata Basin, which is flat terrain at low altitude (Carril
1
Center for Climate and Resilience Research, University et al. 2012; Solman 2013; Solman et al. 2013; Coppola et al.
of Chile, Santiago, Chile
2014). In general, the finest spatial resolution used in these
2
Department of Geophysics, University of Chile, Santiago, long-term RCM-based climate simulations is around 50 km.
Chile

13
Vol.:(0123456789)
 D. Bozkurt et al.

In connection with this, some of the systematic biases such (Bozkurt et al. 2016; Comin et al. 2018), and urbanization
as precipitation overestimation over the mountain ranges and air pollution aspects (Gallardo et al. 2002; Schmitz
were associated with the lack of resolving topographical 2005; Saide et al. 2011; Mazzeo et al. 2018). Although
forcing (Rojas 2006). Furthermore, using the CLARIS-LPB numerical simulations in much of these studies were per-
simulations, Solman (2013) emphasized that the horizon- formed at high spatial resolution (a few km to 15 km)
tal resolution in the current climate simulations over South configurations, they cover a short period (e.g., days to
America is still too coarse to capture some regional features, months). In addition to these case studies, Lenaerts et al.
and increased resolution may lead to better model perfor- (2014) used a high-resolution (5.5 km) RCM driven by
mance, particularly over complex terrain. On the other hand, reanalysis to describe the present-day (1979–2012) cli-
Rojas (2006) showed that higher resolution RCM simula- mate of Patagonia, and they showed that the RCM at 5.5
tions might still include precipitation overestimation over km is able to simulate the orographic uplift and precipita-
the southern Andes due to several factors such as biases in tion generation, capturing the sharp climate gradients in
simulated water vapor content and topography, and physical the southern Andes. On the other hand, in addition to the
configuration of the model. high-resolution simulations driven by reanalysis, GCM-
Continental Chile, lying between the subtropical south- driven RCM simulations for the historical period should
eastern Pacific and the Andes Cordillera, extends from also be evaluated to fully assess the ability of RCMs, par-
18◦ S to 53◦ S and features an extremely complex terrain and ticularly to determine the degree of added value by means
diverse climate zones, which makes it difficult to resolve of resolving potential GCM errors through high-resolution
local-scale features in coarse- and medium-resolution cli- topography and land-surface feedback. Only very recently,
mate model simulations. For instance, the Andes Cordillera, Falco et al. (2018) performed a comprehensive analysis of
has a great influence on local-scale climate variability by RCM evaluation together with the added value analysis at
acting as a climatic wall that disturbs airflow and moisture continental scale of South America based on the available
conditions on both sides of the barrier (Garreaud 2009). In CORDEX simulations at 50 km spatial resolution. In their
an observational study, Viale and Garreaud (2015) showed added value analysis, Falco et al. (2018) concluded that
that precipitation enhancement by orographic uplift on the results of the added value analysis of historical simula-
windward side of the Andes produces large annual precipita- tions with RCMs driven by CMIP5 models are inconclu-
tion (> 2000 mm y­ ear−1) in central-southern Chile. Further sive since added value is only found in certain regions and
south in Chilean Patagonia ( ∼ 40◦ S–55◦ S ), the annual mean in limited simulations.
precipitation reaches 5000–10000 mm in the Patagonian Ice Given the general lack of RCM applications for model
Fields due to the temperate, hyper-humid oceanic climate evaluation as well as added value analysis done with the
and orographic enhancement (Garreaud et al. 2013). On GCM driven simulations, particularly at local scales, the
the other extreme of the country, northern Chile hosts the first aim of this study is to evaluate high-resolution (10 km)
driest desert in the planet with a long-term mean annual hindcast simulations performed by an RCM (RegCM4)
precipitation typically below 5 mm in coastal areas (Mid- (Giorgi et  al. 2012) forced by reanalysis for the period
dleton 2003; Garreaud et al. 2010). Rutllant et al. (2003) 1980–2015. The second aim of this paper is to determine the
highlighted local-scale features such as the steep coastal degree of added value by using historical GCM simulations
topography, the adjacent cold ocean, and local circulations (1975–2005) with the same RCM configuration used in the
forced by the heating of the Andean slopes, as one of the hindcast simulation. To our knowledge, this is the first time
main physical mechanisms that explain Atacama hyperarid- a comprehensive RCM evaluation effort is presented based
ity. In addition to local-scale features, large-scale circulation on 10 km simulations focusing on the Pacific coast and the
patterns such as El Niño-Southern Oscillation (ENSO) and Andes Cordillera of extratropical South America. Further-
Southern Annular Mode (SAM) play a key role on intrasea- more, by taking the advantage of using nested domain tech-
sonal to interdecadal climate variability of Chile (e.g., Rutl- nique with two different spatial resolutions of 0.44◦ ( ∼ 50
lant and Fuenzalida 1991; Silvestri and Vera 2009; Boisier km) and 0.09◦ ( ∼ 10 km), we also analyze the impact of
et al. 2018). resolution on RCM performance in representing present cli-
To understand the interplay between the local- and mate conditions as well as on added value analysis. By doing
large-scale climate in South America and Chile, a number this, we aim to establish confidence on the use of hindcast
of studies have been conducted on regional weather and simulations and dynamically downscaled simulations using
climate model simulations. These studies mainly consist of GCM projections.
dynamical downscaling of boundary conditions provided In Sect. 2, we describe the experiment design, data and
by reanalysis and cover a broad of spectrum of applica- methodology. Section 3 presents the results of the model
tions such as diagnostic studies (Viale et al. 2013; Gar- performance evaluation as well as the added value analy-
reaud et al. 2016), extreme hydrometeorological events sis through comparison of model simulations with the

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

observations. Section 4 summarizes the results and presents (Kiehl et al. 1996); (4) the ocean flux parameterization of
concluding remarks. Zeng et al. (1998); and the scheme of Pal et al. (2000) for
representing resolvable precipitation. Based on a couple of
test simulations done with different convective schemes (not
2 Model description, data and methodology shown), we employed the Grell scheme (Grell 1993) with
a cumulus closure scheme of Fritsch and Chappell (Fritsch
2.1 Model description and experiment design and Chappell 1980). More detailed description of the model
and physical parameterizations can be found in Giorgi et al.
In this study, we employed Regional Climate Model, ver- (2012).
sion 4 (RegCM4), a three-dimensional, primitive equa- The modeling experiment consists of two nested domains
tion, hydrostatic regional climate model developed by the at 0.44◦ ( ∼ 50 km) and 0.09◦ ( ∼ 10 km ) spatial resolutions
International Centre for Theoretical Physics. RegCM4 and 23 vertical levels with a one-way nesting approach
was originally developed by Giorgi et al. (1993a, b) and (Fig.  1). The mother domain covers the whole South
its dynamical core is based on the hydrostatic version of America continent and has 192 × 202 grid cells. The inner
the Penn State/NCAR mesoscale model MM5 (Grell et al. domain covers Chile, part of Argentina and the adjacent
1994). RegCM has been used in numerous regional climate oceans, and has 320 × 520 grid cells based on a Rotated
model simulations, even over regions dominated by com- Mercator projection. For the hindcast simulations, initial and
plex terrain features such as the Atacama Desert (Bozkurt lateral boundary conditions for the mother domain were pro-
et al. 2016), polar regions (Grassi et al. 2013; Bozkurt et al. vided by the European Centre for Medium-Range Weather
2018b), and European Alps (Giorgi et al. 2016). Multiple Forecasts (ECMWF) Reanalysis (ERA-Interim) dataset at
physical schemes are available in the model. In the present 6-h intervals with a grid spacing of 0.75◦ × 0.75◦ (Dee et al.
study RegCM4 runs are performed using (1) the land surface 2011). The nested domain simulation (hereafter RE10) was
model Biosphere-Atmosphere Transfer Scheme (BATS) of then forced by the 3D atmospheric outputs of the mother
Dickinson et al. (1993); (2) the planetary boundary layer of domain (hereafter RE50). Twelve grid points in each direc-
Holtslag et al. (1990); (3) the radiative scheme of the NCAR tion were allocated for each lateral buffer zone in which the
Community Climate System Model Version 3 (CCSM3) models’ prognostic variables were nudged to the boundary

(a) (b)

Fig. 1  a Mother and nested domain topography at 50 km and 10 km resolutions, respectively. b Three sub-regions used in this study. Dashed
black, red and yellow borders identify the Atacama Desert, Central Valley and Patagonian Ice Fields, respectively

13
 D. Bozkurt et al.

conditions with an exponential nudging coefficient proposed that reason, a high-resolution (CR2MET, 0.05◦ resolution)
by Giorgi et al. (1993b). ERA-Interim sea surface tempera- gridded meteorological dataset based on much denser
ture (SST) fields (6 hourly, 0.75◦ × 0.75◦ ) were used as sur- station networks is used as the main reference dataset
face boundary conditions. The simulations were performed to evaluate RegCM4 performance and to perform added
continuously from 1 January 1979 to 31 December 2015. value analysis. CR2MET dataset contain gridded mete-
First year of simulations (1979) was selected as spin-up orological information of precipitation and temperature
period, and thus, not considered in the analysis. (mean, maximum and minimum) for the territory of con-
In addition to the hindcast simulations, we performed tinental Chile covering the period 1979–2016. In terms of
historical simulations covering a period of 1976–2005 with precipitation, in addition to station data information, the
lateral boundary conditions derived from MPI-ESM-MR, CR2MET product is partly based on a statistical downs-
the Earth System Model of the Max Planck Institute for caling of ERA-Interim reanalysis. This approach, which is
Meteorology at medium resolution. MPI-ESM-MR couples also based on station data information, includes multiple
the atmosphere, ocean, sea-ice, and land and was used in a linear regression models to transfer precipitation, moisture
CMIP5-close configuration with spatial and vertical resolu- fluxes and other variables from ERA-Interim onto 0.05◦
tions of T63 ( ∼ 220 km) and L95, respectively (Giorgetta resolution precipitation estimates. The approach considers
2013). The same nested domain setup and physical configu- also local topography and is defined by a set of calibrated
ration of the hindcast simulations were used in the historical parameters with local rainfall observations. Regarding
simulations (hereafter RM50 for mother domain, RM10 for temperature, daily maximum and minimum temperatures
nested domain). RM50 and RM10 were then used to per- were mapped onto the same grid resolution of 0.05◦ using
form the added value analysis [see Sect. 2.3] for the period land surface temperature (LST) estimates from MODIS
1980–2005. satellite retrievals, in addition to near-surface temperature
Summary of the simulations used in this study is given provided by ERA-Interim. Long-term maximum and mini-
in Table 1. We mention that the selection criteria of the mum temperatures were used to adjust the ERA-Interim
MPI-ESM-MR model within CMIP5 models is based on a 3-hourly near-surface temperature so that the adjusted
comprehensive metric evaluation including large-scale cir- 3-hourly data were then used to derive the mean tempera-
culation patterns, SST, ENSO and SAM variability in order ture. More detailed information about the CR2MET can be
to maximize correlations between these large-scale modes found in DGA (2017) and Alvarez-Garreton et al. (2018).
and the observed climate variability over Chile. Detailed The observed dataset used in this study is given in Table 1.
information on selection criteria is given in supplementary In addition to the CR2MET, a catchment dataset of Chile
material (Text S1, Fig. S1). (CAMELS-CL) (Alvarez-Garreton et al. 2018) is used to
further assess the model evaluation by contrasting precipi-
2.2 Data tation and runoff.
The simulation results were also compared with the
Surface station data is provided by the Chilean National driving fields from ERA-Interim. Evaporation product of
Weather Service (Dirección Meteorológica de Chile, GLEAM v3.2a (Martens et al. 2017) is used as a reference to
DMC). A quality control and gap-filling procedures were evaluate the simulated evaporation over the northern Chile.
applied to monthly station data [see Boisier et al. (2016)]. Furthermore, energy balance components from CERES-
Due to the diverse climate features and high spatial cli- EBAF-Surface v4 (Kato et al. 2013) are used as a reference
matic variability in Chile, the commonly used reference dataset to further evaluate model performance over the Ata-
gridded products such as CRU are usually inadequate. For cama Desert.

Table 1  List of the simulations Name Forcing Period Spatial resolution

and observed dataset used in
this study Hindcast
 RE50 ERA-Interim Reanalysis 1979–2015 0.44 × 0.44◦ ( ∼ 50 km)
 RE10 RE50 1979–2015 0.09 × 0.09◦ ( ∼ 10 km)
Historical
 RM50 MPI-ESM-MR 1975–2005 0.44 × 0.44◦ ( ∼ 50 km)
 RM10 RM50 1975–2005 0.09 × 0.09◦ ( ∼ 10 km)
Observation and gridded products
 DMC-stations 1980–2015
 CR2MET 1980–2015 0.05 × 0.05◦ ( ∼ 5 km)
 ERA-Interim 1980–2015 0.75 × 0.75◦ ( ∼ 80 km)

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

2.3 Methodology 0.65 ◦ C∕100 m , using the CR2MET topography as refer-

ence, after regridding onto the common grid. In terms of
Evaluation of RegCM4 model driven by ERA-Interim precipitation, we used first order conservative remapping
was firstly facilitated by using pressure level and surface method that conserves the total area integral of the precipita-
variables. The simulation results were compared with driv- tion (Jones 1999). This technique is assumed to be appropri-
ing fields from ERA-Interim to show the capability of the ate for analysis of precipitation fluxes at different resolution
model to represent surface and pressure level characteristics grids (e.g., Di Luca et al. 2013; Guttler et al. 2015). No
in terms of mean sea level pressure (MSLP), 500 hPa geo- attempt was made to correct the precipitation based on the
potential height ( z500 ), 850 hPa wind vectors and specific elevation differences.
humidity. Then, near-surface air temperature and precipita- A further analysis of the simulations was performed using
tion fields were assessed using the available surface station the added value metric, which shows the squared error dif-
data and gridded products. Mean biases, temporal correla- ferences between the lateral boundary conditions provided
tion, and root mean square difference (RMSD) were used to by MPI-ESM-MR and the downscaled RegCM4 output. It is
evaluate the performance of RegCM4. Evaluation was also calculated as follows (adopting from Di Luca et al. (2016)):
performed in terms of annual cycle and interannual vari-
ability of temperature and precipitation. Note that due to the (XGCM − XOBS )2 − (XRCM − XOBS )2
diverse climate features, we used three subregions to look
AV = ( ) (1)
Max (XGCM − XOBS )2 , (XRCM − XOBS )2
into the performance of the model: Northern Chile, Cen-
tral Chile, and Southern Chile (see in Fig. 1b). Shapefiles where XGCM is the boundary conditions from MPI-ESM-
of each region were used for masking, and only grids over MR, XOBS is the observed reference dataset (CR2MET),
the land areas were used for spatial-average comparisons. and XRCM is the downscaled RegCM4 output driven by
CR2MET was used as the reference observed dataset in MPI-ESM-MR. X in this study corresponds to the temporal
these comparisons. mean (1980–2005) of temperature and precipitation on an
Model evaluation was also performed using the daily annual basis. Positive values (AV > 0) indicate the grids
temperature and precipitation extreme indices following that show added value provided by dynamical downscaling
the definition of the expert team on climate change detec- with RegCM4.
tion and indices (ETCCDI, Karl et al. 1999; Peterson et al. As all variables must be on the same grid for added value,
2001). These include 90th percentile of daily maximum tem- we used three representative grid resolutions of 0.44◦ , 1◦
perature (TX90), frost days (FD), 95th percentile of daily and 1.5◦ resolutions. In order to minimize uncertainties and
precipitation at wet days (R95), and consecutive dry days possible errors arising from regridding, particularly when
(CDD). All indices were calculated using the data for the upscaling from high resolution dataset of CR2MET and
entire year. Simulated temperature and precipitation indices RM10, we gradually regridded each dataset following the
of RE50 and RE10 were assessed against the CR2MET data- same methodology performed by Torma et al. (2015). In
set. Detailed definitions of the indices used in this study are other words, three common grids are used as intermediate
listed in Table 2. A full descriptive list of the indices can be resolutions, and for instance, CR2MET and RM10 were
found on the ETCCDI website http://etccd​i.pacif​i ccli​mate. upscaled to the multiple low resolutions ( 0.44◦ , 1◦ , and 1.5◦
org/list_27_indic​es.shtml​. resolutions) step by step. We used bilinear and first order
To ease comparison between model simulations and conservative remapping for temperature and precipitation,
observations, we used 0.44◦ resolution as a common grid, respectively. We applied the same approach of elevation
therefore, high resolution dataset of CR2MET and RE10 correction for temperature fields of RM10, RM50 and MPI-
were upscaled to the 0.44◦ resolution. In order to have the ESM-MR fields after regridding. Given that we are using
exact grid numbers, temperature fields from each data- three different sub-regions of Chile in a narrow area charac-
set including RE50 and ERA-Interim too, were interpo- terized by complex terrain, 1.5◦ grid resolution does not have
lated onto the 0.44◦ resolution using bilinear interpola- enough number of grids in the sub-regions to assess and
tion. Additionally, an elevation correction was carried out quantify the simulation results. Therefore, we only provide
for temperature assuming a uniform temperature lapse of results from 0.44◦ and 1◦ grid resolutions.
Table 2  Definitions of extreme Indice label Indice name Indice definition Units
temperature and precipitation
indices used in this study TX90 Extreme temperature 90th percentile of daily maximum temperature ◦
C
FD Frost days Number of days where daily minimum temperature < 0 ◦C days
R95 Extreme precipitation 95th percentile of daily precipitation on wet days (R ≥ 1 mm) mm
CDD Consecutive dry days Maximum number of consecutive dry days where R < 1 mm days

13
 D. Bozkurt et al.

3 Results (− 2 to − 5 ◦ C ) (Figs. 2d, S4). Compared to the RE50, RE10

generally reduces the cold bias over the Central Valley and
In this section, we first present the performance evaluation of Cordillera due to the finer topography. On the other hand,
RegCM4 model driven by ERA-Interim in terms of surface RE10 overestimates temperature in northern Chile, par-
variables of temperature and precipitation (Sect. 3.1). We ticularly in Atacama Desert ( ∼ 4 ◦ C ) (Fig. 2e, S4), while
then present the added value analysis using the MPI-ESM- it tends to simulate colder temperatures over the southern
MR, RM10 and RM50 results (Sect. 3.2). Chile. RE50 represents a closer estimation of observed
temperature over the Atacama Desert. However, this does
3.1 Performance of RegCM4 not imply that RE50 has a better representation of the sur-
face fluxes but rather it may include more systematic errors
An evaluation of the large-scale circulation based on synop- leading to wrong characterization of the temperature pattern
tic fields is provided in full detail in the supplementary mate- over the desert. Indeed, RE10 gives a closer estimation and
rial (Text S2, Figs. S2, S3). Overall, RegCM4 is capable of variability of evaporation to that in the reference dataset of
reproducing mean spatial fields of important large-scale fea- GLEAMv3.2a over the Atacama Desert (Fig. S5a).
tures such as the Southeast Pacific Subtropical Anticyclone The Atacama warm bias in RE10 is further investigated
(SPSA), the mid-latitude westerly winds over the Patago- using surface energy balance components. Figure 3 shows
nia and low-level moisture distribution along both sides of mean annual averaged of surface energy balance compo-
the Andes barrier. The higher resolution simulation (RE10) nents ( W m2−1 ) over the Atacama Desert (see Fig. 1b) from
generally gives a better performance in terms of represent- CERES-EBAF-v4, RE50 and RE10. Both simulations have
ing magnitude of these fields compared to those in RE50 at almost the same amount of net absorbed solar radiation at
annual and summer time scales. Both simulations tend to the surface with an underestimation compared to the esti-
have stronger westerlies compared to the ERA-Interim over mated values of CERES-EBAF-v4 over the Atacama Desert.
the Patagonia. The warm surface temperature bias of RE10 is expected to
lead to an overestimation of upwelling longwave radiation,
3.1.1 Temperature regimes however, it is underestimated compared to CERES-EBAF-
v4 dataset. An overestimation exists in the simulated down-
Figure 2 shows the mean observed, ERA-Interim and simu- welling longwave radiation for RE10. Although this can be a
lated annual temperature distribution over the region for the consequence of the warm surface temperature bias, it could
period of 1980–2015. Surface station data indicate a dis- be argued that the overestimation of downwelling longwave
tinct latitudinal distribution of the temperature with values radiation and associated underestimation of net radiative
lower than 10 ◦ C in the southern Chile ( 42 ◦ S and 55 ◦ S ) cooling in RE10 compared to the CERES-EBAF-v4 and
and 16 ◦ C to 22 ◦ C in central and northern Chile (Fig. 2a). RE50, can partially control the warm surface temperature
Zonal temperature contrast also exists, particularly over the bias. Furthermore, RE10 tends to have larger surface specific
Altiplano sector in northern Chile with a temperature dif- humidity values compared to those in the RE50 indicating
ference around 10 ◦ C between the coastal areas and higher a larger greenhouse effect locally (Fig. S5b). These discrep-
elevations. These spatial features are represented reasonably ancies might be associated with lack of dust module and
well in the gridded product of the CR2MET with finer local accurate land surface processes such as surface emissivity
details of temperature such as Patagonian Ice Fields in the and albedo in the model configuration used in this study.
south and Salar de Atacama in the north (Fig. 2b). ERA- Indeed, for instance, Marcella and Eltahir (2012) corrected
Interim yields a consistent temperature pattern with the a warm bias in RegCM3 simulations over semi-arid south-
CR2MET over Chile, yet it exhibit higher (+ 1 to + 5 ◦ C ) and west Asia by modifying model’s surface albedo values and
lower (− 1 to − 5 ◦ C ) temperature ranges along the north- incorporating RegCM3’s dust module.
ern Chile coasts and over the Cordillera of central-northern Figure 4 presents the mean annual cycle and interan-
Chile, respectively (Fig. 2c, also see in Fig. S4). nual variability of temperature (1980–2015 period) from
Overall, the downscaled simulations of ERA-Interim CR2MET, ERA-Interim and simulations for three subre-
at two different spatial resolutions capture well the spatial gions. Overall, both simulations at different resolutions illus-
pattern of mean annual temperature. Nonetheless, there are trate a very close spatial-mean temperature climatology and
noticeable differences between the simulated and observed variability in each subregion. However, some regional differ-
temperatures. Temperature distribution in RE50 illustrates ences exist among the simulations. For instance, in northern
a similar pattern to that in ERA-Interim, however, RE50 Chile, RE50 shows an underestimation of temperature, while
underestimates temperature over the central-southern Chile, RE10 is characterized by a warm bias (close to + 1 ◦ C ) from
particularly over the Cordillera of central-northern Chile January to March (Fig. 4a). Regarding the interannual vari-
ability of temperature, both simulations show a moderate

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

Fig. 2  a Spatial distribution (a) (b)

of 36-year (1980–2015) mean
annual average temperature
( ◦ C ) from surface observa-
tions, b gridded observation
(CR2MET), c ERA-Interim,
d dynamically downscaled
simulations of ERA-Interim
at 50-km (RE50) and e 10-km
(RE10) resolutions

(c) (d) (e)

temporal correlation with CR2MET (0.42 for RE50 and 0.45 related to the recent cooling trend in coastal SST (Falvey
for RE10) in northern Chile. RE10 has a better representa- and Garreaud 2009). Falvey and Garreaud (2009) showed
tion of temperature variability with lower bias and RMSD in a recent cooling trend ( − 0.2 ◦ C decade−1 ) off the coast
central Chile. In southern Chile, both simulations have the of northern and central Chile for the period 1979–2006.
highest temporal correlation among the three sub-regions They also highlighted a strong contrast between surface
(0.82 for RE50 and 0.83 for RE10), yet a systematic cold cooling at coastal stations and warming in the Andes
bias exist in both simulations with larger RMSD values in ( + 0.25 ◦ C decade−1 ). Vuille et al. (2015) also illustrated
RE10 the cooling along the coastal zone and warming in the
Large discrepancies between the observed and simu- Andes using station data. A trend map of observed and
lated temperature variability in northern Chile could be simulated temperature for the period 1980–2015 shows

13
 D. Bozkurt et al.

Fig. 3  Surface energy bal-

ance components ( W m2−1 )
over the Atacama Desert (see
Fig. 1b) from CERES-EBAF-v4
(gray), dynamically downscaled
simulations of ERA-Interim at
50-km (RE50, red) and 10-km
(RE10, blue) resolutions. Values
correspond to mean annual
average for the period 2001-
2015 for CERES-EBAF-v4,
1980-2015 for RE50 and RE10.
Snet : Net shortwave radiation,
Ldown : Downward longwave
radiation, Lupw : Upward long-
wave radiation, Lnet : Net long-
wave radiation, Qtur : Turbulent
heat fluxes (i.e., sensible and
latent heat fluxes)

that both RCM simulations capture the cooling trend as well as dynamically downscaled simulations are given
( ∼ − 0.1 to − 0.3 ◦ C decade−1 ) in the coastal area, how- in Fig. 7. Rain-gauge data shows the striking precipitation
ever, RE50 fails to capture the warming trend over most gradient increasing from north ( < 100 mm) to south ( > 600
parts of the Andes while RE10 captures some observed mm) as well as from coastal areas ( ∼ 1000 mm) to the Andes
warming trend over the central Andes ( ∼ 28◦ S to ∼ 35◦ S ) ( ∼ 4000 mm) in central-southern Chile (Fig. 7a). CR2MET
(Fig. 5c, d). represents well the station-based spatial pattern of precipi-
Difficulties in capturing the temperature variability tation including north to south precipitation increase and
in northern Chile result in poor performance of repre- orographic modification of the Andes in central-southern
senting TX90 variability (e.g., very low correlations) Chile (Fig. 7b).
for both simulations over this region (Fig. 6). Particu- ERA-Interim is characterized by the same north to south
larly, a systematic warm bias is clear in RE10. In central precipitation gradient with a slightly different magnitudes
Chile, both simulations largely capture the interannual such as more precipitation in Patagonia and northern
variability of TX90, however, RE50 exhibits a system- Chile (Fig. 7c). RE50 does not capture well the orographic
atic underestimation ( ∼ 2 ◦ C ). RE10 tends to have larger enhancement of precipitation and shows a dry bias over the
than observed TX90 on the broad areas of central Chile mountains areas of central-southern Chile (Figs. 7d, S7).
and Central Valley, and lower than observed TX90 over Wetter conditions in the Central Valley zone and Andes
the Andes ranges as well as the Patagonian Ice Fields. A ranges of northern Chile are also notable in RE50. On the
similar comparison of temperature extremes shows that other hand, RE10 captures the main features of the clima-
both simulations capture well the FD variability, yet RE10 tological pattern of precipitation such as latitudinal varia-
exhibits a better spatial and temporal variability of the FD tion and orographic forcing. Particularly, RE10 enhances
over the southern Chile where the Patagonian Ice Fields orographic forcing and reproduces precipitation increase
are located (Fig. S6). reasonably well in western slope of the Andes in central-
southern Chile (Fig. 7e). Compared to its forcing field of
3.1.2 Precipitation regimes ERA-Interim, improvement of orographic precipitation
in RE10 can also be seen in the bias map with respect to
Long-term mean (1980–2015) annual accumulated precipi- CR2MET (Fig. S7). On the other hand, RE10 tends to have a
tation from the observed station data and gridded products large standard deviation of annual accumulated precipitation

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

CR2MET RE50
ERA-INT RE10

(a)
r: 0.45, 0.42
RMSD: 0.46, 1.21 (oC)
Bias: -0.05, -1.11 (oC)

(b)
r: 0.66, 0.63
RMSD: 0.95, 1.54 (oC)
Bias: -0.91, -1.51 (oC)

(c) r: 0.83, 0.82

RMSD: 1.03, 0.89 (oC)
Bias: -1.00, -0.85 (oC)

Fig. 4  a 36-year (1980–2015) mean annual cycle and inter-annual and the dashed lines correspond to ERA-Interim (gray), dynamically
variability of temperature ( ◦ C ) for Northern Chile, b Central Chile, downscaled simulations of ERA-Interim at 50-km (RE50, red) and
c Southern Chile. The solid line corresponds to CR2MET (black), 10-km (RE10, blue) resolutions

( ∼ 1000 mm) on some grids within the Andes ranges of cen- El Niño and La Niño years) are well reproduced by both
tral Chile and Patagonian Ice Fields (Fig. S8a). Both simula- simulations indicating no significant performance improve-
tions, however, capture well the spatial pattern of coefficient ment with higher resolution.
of variation with the largest values (0.6–0.9) in northern In recent years, central Chile has been experiencing a
Chile (Fig. S8b). persistent deficit of precipitation ( ∼ 30%) (Boisier et al.
The mean annual cycle and interannual variability of 2016), which is expected to continue in the future based on
precipitation (1980–2015 period) for three sub-regions are the climate projections (Demaria et al. 2013; Bozkurt et al.
given in Fig.  8. In northern Chile, overall, both simula- 2018a). This so-called mega-drought has caused significant
tions capture the observed annual cycle shape and interan- impacts in surface hydrology and vegetation productivity in
nual variability with a systematic overestimation. RE50 central Chile (Garreaud et al. 2017). A further comparison
substantially overestimates the precipitation, on the other of the precipitation anomaly during the 2010–2015 period
hand, RE10 closely agrees with the observed values and with respect to long-term mean shows that both simula-
ERA-Interim, indicating the improvement of the model per- tions reasonably produce the spatial extent of the recent
formance at finer resolution. In central Chile, spatial-mean drying period ( ∼ 30◦ S to ∼ 40◦ S ) (Fig. S9). Long-term
precipitation climatology and interannual variability (e.g., (1980–2015) trend map of precipitation exhibits that the

13
 D. Bozkurt et al.

Fig. 5  (a) Trends in annual (a)

temperature ( ◦ C 10 yrs−1 )
for the period 1980–2015 for
CR2MET, (b) ERA-Interim,
(c) dynamically downscaled
simulations of ERA-Interim at
50-km (RE50) and (d) 10-km
(RE10) resolutions

(b) (c) (d)

simulations, especially RE10, also capture the long-standing and 45◦ S ) is presented in Fig. 9. Overall, 10 km simula-
drying trend in central-southern Chile showed by Boisier tion results exhibit a better representation of precipitation
et al. (2018) (Fig. S10). variability and suppress an overestimation of precipitation
A well-distributed pattern of precipitation along the year over the Andes Cordillera of northern Chile found in the
is captured by both simulations in southern Chile with a 50 km simulation. Furthermore, high-resolution simula-
slight systematic underestimation. Both simulations success- tions capture well the zonal precipitation variability over
fully reproduce most of the interannual variability of pre- the coastal range and the upstream of the Andes range in
cipitation in southern Chile, with a slightly better temporal central-southern Chile.
correlation in RE10 (0.76) than that in RE50 (0.72). A further comparison was performed by means of run-
To take a closer look at the simulated precipitation com- off-precipitation covariability. Figure 10a presents the scat-
parison, zonal variation of mean annual accumulated pre- ter diagram of long-term mean (1980–2015) precipitation
cipitation along four different latitudes ( 30◦ S , 35◦ S , 40◦ S from CR2MET versus observed runoff from CAMELS-CL

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

(a) (b)

Fig. 6  (a) Spatial distribution of 36-year (1980–2015) mean annual Inter-annual variability of TX90 for Northern Chile and Central
TX90 (90th percentile of the daily maximum temperature, ◦ C ) from Chile. The solid line corresponds to CR2MET (black), and the dashed
CR2MET, ERA-Interim, dynamically downscaled simulations of lines correspond to ERA-Interim (gray), RE50 (red) and RE10 (blue)
ERA-Interim at 50-km (RE50) and 10-km (RE10) resolutions. (b)

in central Chile. Overall, CR2MET reveals a close fit to the interannual covariability, CR2MET exhibits R2 values larger
1:1 line in central Chile ( R2 = 0.87 ) with runoff ratios (Run- than 0.6 in central Chile. RE50 shows a poorer performance
off/Precipitation) ranging between expected values (0.5 and with low R2 values ( < 0.5) while RE10 has a slightly better
0.8) (Alvarez-Garreton et al. 2018). However, there are some performance in representing the runoff-precipitation interan-
values with runoff ratios greater than 1 that indicate a precip- nual covariability.
itation underestimation assuming that streamflow data and Finally we close this subsection with precipitation
catchment area are reliable (Alvarez-Garreton et al. 2018). extremes. Long-term mean (1980–2015) R95 values from
RE50 gives relatively low runoff ratios (i.e., precipitation the observation, ERA-Interim and simulations are given in
overestimation, see also Fig. S7) with a lower coefficient of Fig. 11a. Observed R95 values of daily precipitation indicate
determination (0.66) (Fig. 10b). RE10 exhibit a very similar that most parts of the Andes along with small spots in the
scatter with CR2MET in which a close fit to the 1:1 line is Central Valley and Patagonian Ice Fields feature high daily
kept (Fig. 10b). Compared to the RE50, RE10 gives a better precipitation values (> 60 mm), and largest values are con-
representation of precipitation-runoff covariability with a centrated around 35◦ S ( ∼ 80 mm) over the Cordillera in cen-
larger R2 value (0.83) indicating an improvement of the sim- tral Chile. RE50 shows high R95 values (50–70 mm) mostly
ulations with the use of higher spatial resolution. Similar to concentrated along the Central Valley zone and it does not
CR2MET, runoff ratios larger than 1 also exist in both simu- represent the observed maximum of R95 over the central
lations indicating the precipitation underestimation, which Andes ranges. On the other hand, RE10 captures well the
was also observed in different precipitation products (Alva- spatial distribution of R95 maxima over the Andes ranges in
rez-Garreton et al. 2018). Regarding the runoff-precipitation

13
 D. Bozkurt et al.

Fig. 7  a Spatial distribution (a) (b)

of 36-year (1980–2015) mean
annual total precipitation
(mm) from surface observa-
tions, b gridded observation
(CR2MET), c ERA-Interim,
d dynamically downscaled
simulations of ERA-Interim
at 50-km (RE50) and e 10-km
(RE10) resolutions

(c) (d) (e)

central Chile albeit notably larger than the observed values is underestimated (wetter conditions) over the Andes range
(> 90 mm). Both simulations show high skill in simulating of central and northern Chile (Fig. S11).
interannual variability of R95 over central Chile represent- Overall, so far, long-term climatological analysis at daily,
ing the observed increasing and decreasing tendency of R95 monthly, annual and seasonal scales show that RE10 has a
during El Niño and La Niña years, respectively (e.g., 1997 better performance than RE50 in preserving the observed
and 1998) (Fig. 11b). However, a systematic overestimation spatial and temporal variabilities, particularly for precipita-
of R95 exists in RE10 mainly due to the larger R95 over tion, despite some discrepancies in temperature variability
the Andes ranges in central Chile. In terms of CDD, both over the northern Chile.
simulation follow closely forcing field of ERA-Interim and,
overall there is a good consistency between the observed and
simulated CDD patterns. However, the magnitude of CDD

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

CR2MET RE50
ERA-INT RE10
(a)
r: 0.68, 0.69
RMSD: 92, 162 (mm)
Bias: +82, +148 (mm)

(b)
r: 0.86, 0.84
RMSD: 221, 252 (mm)
Bias: -152, +168 (mm)

(c) r: 0.76, 0.72

RMSD: 126, 196 (mm)
Bias: -77, -163 (mm)

Fig. 8  a 36-year (1980–2015) mean annual cycle and inter-annual and the dashed lines correspond to ERA-Interim (gray), dynamically
variability of precipitation (mm) for Northern Chile, b Central Chile, downscaled simulations of ERA-Interim at 50-km (RE50, red) and
c Southern Chile. The solid line corresponds to CR2MET (black), 10-km (RE10, blue) resolutions

3.2 Added value analysis temperature distribution by capturing local scale features

such as warm temperatures over the Atacama Desert and
3.2.1 Temperature east-west gradient (Fig. 12d).
A more quantitative analysis of added value for the both
In this subsection, we first compare the spatial pattern of common grids of 0.44◦ and 1◦ indicate that there is a clear
annual mean temperature (1980–2005) from MPI-ESM-MR evidence of added value for temperature over the northern
and dynamically downscaled simulations forced with MPI- and central Chile (Fig. 13a). However, simulated tempera-
ESM-MR to those in the gridded observations (Fig. 12). ture by RM10 adds more value by contrasting the features
Overall, compared to CR2MET, MPI-ESM-MR shows to between central zone and the Andes ranges in northern and
some degree the north to south temperature gradient, how- central Chile. Indeed, the ratio of amount of added value
ever, due to the coarse resolution, it misses the east-west gra- points to total points inside of each sub-region indicates
dient imposed by the topography (Fig. 12b). Furthermore, it larger values for RM10 in northern and central Chile. Fur-
tends to underestimate temperature over the central parts of thermore, RM50 tends to not have added value in simulating
northern and central Chile. Compared to the MPI-ESM-MR temperature distribution over the large parts of the Central
and RM50, RM10 represents a closer estimation of observed Valley and the Cordillera of central-northern Chile. When

13
 D. Bozkurt et al.

Fig. 9  36-year (1980–2015) mean annual zonal precipitation (mm) (gray), dynamically downscaled simulations of ERA-Interim at
along 30◦ S , 35◦ S , 40◦ S and 45◦ S . The solid line corresponds to 50-km (RE50, red) and 10-km (RE10, blue) resolutions
CR2MET (black), and the dashed lines correspond to ERA-Interim

using the 1◦ resolution, upscaled RM10 results again indicate RM10 illustrates more added value and thus, it has a
that there is more evidence of added value for temperature larger ratio of amount of added value points to total points
over the northern and central Chile compared to the RM50 in northern and central Chile with a marked improvement of
(Fig. 13b). Compared to the northern and central parts, both precipitation distribution along the coastal range (Fig. 15a).
simulations show less added value over the southern Chile. However, both simulations miss added value over the Andes
It should also be noted that upscaled RM10 at a coarser ranges in northern Chile indicating the difficulties improv-
resolution close to the MPI-ESM-MR resolution ( 1.5◦ ) still ing the driving MPI-ESM-MR over these regions. Although
exhibits added value in northern and central-southern Chile RM50 shows significant added value over the central-south-
(Fig. S12a). ern Andes range and most parts of northern Patagonia, it has
no added value over large parts of central Chile mostly along
3.2.2 Precipitation the coast and Central Valley zone. On the other hand, RM10
indicates added value due to the increased resolution along
Compared to the MPI-ESM-MR and RM50, RM10 gives a the coastal range as well as over the Andes range in central-
closer estimation of observed precipitation distribution with southern Chile, highlighting the impact of the enhanced local
a better representation of orographic precipitation, although information on the quality of the simulations. Similar spatial
a wet bias exists over the Andes ranges, particularly in cen- extent of added value exists when upscaling to the coarser
tral Chile and northern Chile (Fig. 14). A high standard common grids of 1◦ and 1.5◦ , with less added value along the
deviation of annual accumulated precipitation ( ∼ 1000 mm) coastal range of central Chile in RM10 (Figs. 15b, S12b).
within the Andes range of central Chile and Patagonian Ice We close this section with a comparison of spatial-mean
Fields is also notable in RM10 (Fig. S13a). Both dynami- daily precipitation intensity distributions from the obser-
cally downscaled simulations, particularly RM10, tend to vation, simulations and the driving MPI-ESM-MR over
improve the MPI-ESM-MR fields in representing the spatial northern and central Chile (Fig. 16). Northern Chile is char-
pattern of coefficient of variation in northern Chile except acterized by frequent low precipitation events (between 1
for the high Andes (Fig. S13b). and 5 mm day−1), and MPI-ESM-MR substantially overes-
timates the number of low precipitation events in northern

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

(a) (b)

RE50 (R2 = 0.66)

RE10 (R2 = 0.83)

(c)

Fig. 10  a 36-year (1980–2015) mean annual precipitation (mm) from (RE10, blue) resolutions. c Zonal distribution of coefficient of deter-
CR2MET against observed mean annual runoff (mm) over central mination calculated from interannual (1980–2015) CR2MET (gray),
Chile. b The same in a but for precipitation from dynamically down- RE50 (red) and RE10 (blue) precipitation fields against observed run-
scaled simulations of ERA-Interim at 50-km (RE50, red) and 10-km off

Chile (Fig. 16a). Both simulations, especially RM50, tend events albeit with a slight overestimation. The results for
to overestimate low precipitation events, however, RM10 extreme precipitation events indicate that both simulations
improves slightly the representation of number low precipi- represents more frequent heavier precipitation events in cen-
tation events. In terms of extreme daily precipitation events, tral Chile than what is found in reality.
RM50 illustrates more frequent heavier precipitation events
(> 15 mm day−1), and in general, RM10 has a slightly better
representation of extreme precipitation events in northern 4 Summary and concluding remarks
Chile. In central Chile, although low and moderate precipita-
tion events are more frequent (between 5 and 20 mm day−1 ) This study evaluates several simulations performed with
heavy precipitation events ( > 45 mm day−1 ) can be also RegCM4 over Chile. The performance of RegCM4 is eval-
observed (Fig. 16b). Overall, a systematic underestimation uated through two nested hindcast simulations forced by
of all types of precipitation events exists in MPI-ESM-MR. ERA-Interim for the period 1980–2015. Added value anal-
The results with RM10, on the other hand, are systematically ysis is also performed using two nested simulations forced
better than those of MPI-ESM-MR and give a close estima- by a global climate model (GCM, MPI-ESM-MR) for the
tion of observed number of low and moderate precipitation period 1980–2005. The performance evaluations are carried

13
 D. Bozkurt et al.

Fig. 11  a Spatial distribution

of 36-year (1980–2015) mean
annual R95 (95th percentile of (a)
the daily precipitation, mm)
from CR2MET, ERA-Interim,
dynamically downscaled
simulations of ERA-Interim
at 50-km (RE50) and 10-km
(RE10) resolutions. b Inter-
annual variability of R95 for
Central Chile. The solid line
corresponds to CR2MET
(black), and the dashed lines
correspond to ERA-Interim
(gray), RE50 (red) and RE10
(blue)

(b)
r: 0.78, 0.72
RMSD: 8.8, 6.2 (mm) CR2MET RE50
Bias: +7.5, +4.9 (mm) ERA-INT RE10

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

Fig. 12  a Spatial distribution

of 26-year (1980–2005) mean (a)
annual average temperature
( ◦ C ) from gridded observation
(CR2MET), b MPI-ESM-MR,
c dynamically downscaled
simulations of MPI-ESM-MR
at 50-km (RM50) and d 10-km
(RM10) resolutions

(b) (c) (d)

out through a comparison of mean large-scale fields over the gridded meteorological observations of Chile (CR2MET) as
Pacific coast and Andes Cordillera as well as surface varia- well as to ERA-Interim. Based on the analyses, the following
bles of temperature and precipitation at annual, monthly and main findings can be highlighted:
daily time scales over three subregions of mainland Chile:
Northern Chile, Central Chile, and Southern Chile. Using • In general, RegCM4 shows good skills in simulating the
nested domains at 0.44◦ ( ∼ 50 km, covering all South Amer- spatial variability of mean annual temperature. However,
ica) and 0.09◦ ( ∼ 10 km, focusing on Chile) allows to analyze there are some differences between the simulated and
the role of resolution in capturing the present climate con- observed temperatures at local scales such as a system-
ditions (for RE50 and RE10) as well as in determining the atic warm bias ( ∼ + 4  ◦ C) over the Atacama Desert in
degree of the added value (for RM50 and RM10). The simu- RE10.
lation results were primarily compared to a high-resolution

13
 D. Bozkurt et al.

(a) the northern and central Chilean coasts, they largely miss
the interannual variability, and RE50 does not reproduce
the observed warming trend over the Andes ranges of
central Chile.
• Difficulties in capturing the temperature variability in
northern Chile result in even poorer representation of
daily temperature extremes (i.e., 90th percentile of maxi-
mum temperature) in both simulations over this region.
On the other hand, RE10 represents a better spatial and
temporal variability of daily temperature extremes (i.e.,
frost days) over the southern Chile where the Patagonian
Ice Fields are located.
• Regarding precipitation, RE10 better represents the cli-
matology and annual cycle compared to the RE50. Fur-
thermore, RE10 reasonably represents orographic uplift
and precipitation generation on the windward side of the
Andes, however, it shows a high amplitude of precipita-
tion variability over the Andes ranges.
• Both simulations capture the spatial-mean interannual
variability of precipitation in each subregion (e.g., ENSO
years and mega-drought in central Chile). RE10 has a
(b) better representation of the zonal precipitation variabil-
ity and for instance, it suppresses the overestimation of
precipitation over the Andes Cordillera of northern Chile
found in RE50.
• Compared to the ERA-Interim and RE50, RE10 repre-
sents reasonably well the spatial distribution daily pre-
cipitation extremes, although with notably larger than
the observed values over the Andes in central Chile.
Both simulations capture the overall spatial variability
of consecutive dry days, however, with a wet bias over
the Andes.
• In general, both simulations add value to those of the
driving MPI-ESM-MR over large parts of Chile. How-
ever, this depends on the subregion and variable, and
both simulation can exhibit different results.
• There is a clear evidence that RM10 adds more value for
both temperature and precipitation in northern and cen-
tral Chile. Both simulations have difficulties in improv-
ing the driving MPI-ESM-MR over the Andes ranges in
Fig. 13  a 26-year (1980–2005) mean annual temperature ( ◦ C ) from
northern Chile. Compared to the northern and central
dynamically downscaled simulations of MPI-ESM-MR at 50-km parts, both simulations show less added value in southern
(RM50) and 10-km (RM10) resolutions. Stippling indicates grid Chile.
points where there is added value by the dynamical downscaling. All • Frequency distribution of daily precipitation over north-
datasets were remapped onto a common grid of 0.44◦ resolution and
an elevation correction was carried out assuming a uniform tempera-
ern and central Chile shows that RM10 adds value to
ture lapse rate of 0.65 ◦ C∕100 m , using the CR2MET topography as the representation of extreme precipitation. For instance,
reference. b The same as a but with a common grid of 1◦ resolution. it has a better performance in presenting the very dry
Numbers correspond to the ratio of amount of added value points to regime of the northern Chile than that in MPI-ESM-MR
total points inside of each sub-region
and RM50 by improving the number of low precipitation
events.
• There are large discrepancies between the simulated and
observed temperatures in northern Chile. Although both Our work has potentially important implications for the
simulations represent the observed coastal cooling along dynamical downscaling efforts over complex terrains

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

Fig. 14  a Spatial distribution (a)

of 26-year (1980–2005) mean
annual total precipitation (mm)
from gridded observation
(CR2MET), b MPI-ESM-MR,
c dynamically downscaled
simulations of MPI-ESM-MR
at 50-km (RM50) and d 10-km
(RM10) resolutions

(b) (c) (d)

featuring diverse climatic gradients such as Chile. Based and improve the representation of temperature and precipi-
on the results of the present study, it could be said that tation over Chile’s complex terrain particularly in central-
coarse resolution GCMs miss important climate gradients southern Chile. Our results also illustrate a clear evidence
imposed by the topography, and more importantly, this fea- of the added value with the use of high-resolution RCMs
ture is shared with reanalysis products as well as medium- suggesting that spatial resolution is critical for a better esti-
resolution RCM simulations ( ∼ 50 km). Overall, high-res- mation of spatial and temporal variability of present climate
olution RCM simulation results ( ∼ 10 km) are promising, conditions. This is in agreement with previous studies over

13
 D. Bozkurt et al.

(a) On the other hand, some large discrepancies between the

observed and simulated climate variabilities such as tem-
perature trend contrast between coastal zone and Andes
ranges of central-northern Chile, and extreme precipitation
inconsistencies along the Andes Cordillera in ERA-Interim
driven simulations need further investigation. These discrep-
ancies might be due not only to model shortcomings and/
or physical configuration, but also to the inadequate obser-
vational network over the regions with complex terrain. For
instance, lack of sufficient and high-quality observational
data over the extreme environments such as the high Andes
Cordillera and Patagonia can lead to a number of potential
uncertainties, and have a large impact on the robustness of
the observational gridded product used as “observed truth”.
Similar difficulties have been highlighted in previous studies
on regions with complex terrain (e.g., Onol 2012; Bozkurt
et al. 2012; Solman 2013; Bieniek et al. 2016). Improving
the quality of observational networks over the complex ter-
rain can also be beneficial in implementing bias correction
techniques in case of existence of persistent systematic errors
in the simulated fields. Regarding modeling technique, our
(b)
study suggests that beside the general improvement of cli-
mate variability via increased resolution, further efforts are
required in improving the dynamical downscaling results.
For instance, especially over the Andes Cordillera and Ata-
cama Desert, model tuning and optimization are important
to improve the capability of the RCM results in represent-
ing climate variability over these regions. A comparison of
regional ocean-atmosphere coupled model to stand-alone
RCM simulations might be needed to better resolve the
impact of coastal features such as marine stratiform clouds
and coastal SST forcing on the uncertainties in the simulated
climate variability in northern Chile.
In conclusion, given that the highest spatial resolution of
the current CORDEX-like initiatives over the South America
is about 50 km and upcoming CMIP6 GCM simulations will
cross the threshold of 50 km resolution, more RCM applica-
tions with higher resolutions are important to improve our
understanding of the dynamical downscaling efforts over
complex terrains and extreme environments. Furthermore,
Fig. 15  a 26-year (1980–2005) mean annual total precipitation (mm) as the quality of the boundary conditions as well as physical
from dynamically downscaled simulations of MPI-ESM-MR at configuration of RCMs play a key role in determining the
50-km (RM50) and 10-km (RM10) resolutions. Stippling indicates robustness of the simulated fields, multi-model high-reso-
grid points where there is added value by the dynamical downscaling.
lution RCMs forced with different GCMs are also needed,
All datasets were remapped onto a common grid of 0.44◦ resolution.
b The same as a but with a common grid of 1◦ resolution. Numbers despite entailing a high computational cost. Development
correspond to the ratio of amount of added value points to total points and continuation of CORDEX-like initiatives over the South
inside of each sub-region America domain with a higher spatial resolution than the
current available simulations are important to perform fur-
ther model evaluation and added value analyses that may
complex terrains in other regions (e.g., Walker and Diffen- be more relevant and beneficial for the generation of more
baugh 2009; Di Luca et al. 2013; Onol 2012; Montesarchio robust high-resolution climate change projections at local-
et al. 2014; Torma et al. 2015; Shi et al. 2018). and regional-scales.

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

Fig. 16  a Histogram of fre- (a)

quency distribution of daily
precipitation (mm, 1980–2005)
for Northern Chile from
CR2MET (black), MPI-ESM-
MR (orange), dynamically
downscaled simulations of MPI-
ESM-MR at 50-km (RM50,
red) and 10-km (RM10, blue)
resolutions. b The same in a but
for Central Chile. The calcula-
tion is based on all grid point
values of the original grids of
each dataset

(b)

Acknowledgements  This work was funded by FONDAP-CONICYT decline and recent megadrought in Central Chile. Geophys Res
15110009. The authors acknowledge the anonymous reviewer for the Lett 43(1):413–421. https​://doi.org/10.1002/2015G​L0672​65
constructive comments on the manuscript. The simulations were per- Boisier JP, Alvarez-Garreton C, Cordero R, Damiani A, Gallardo L,
formed within a project entitled “Simulaciones climáticas regionales y Garreaud R, Lambert F, Ramallo C, Rojas M, Rondanelli R (2018)
marco de evaluación de la vulnerabilidad” funded by Chilean Ministry Anthropogenic drying in Central-Southern Chile evidenced by
of Environment. A platform has been developed within that project and long-term observations and climate model simulations. Elem Sci
all the simulation outputs described in this study can be accessible from Anth. https​://doi.org/10.1525/eleme​nta.328
that platform following the CORDEX data format protocols, available Bozkurt D, Turuncoglu U, Sen OL, Dalfes HN (2012) Downscaled
at http://simul​acion​es.cr2.cl/. The authors appreciate the support from simulations of the ECHAM5, CCSM3 and HadCM3 global mod-
Francisca Muñoz and Nancy Valdebenito at the Data and Computing els for the Eastern Mediterranean-Black Sea region: evaluation
unit at (CR)2. Powered@NLHPC: This research was supported by the of the reference period. Clim Dyn 39(1–2):207–225. https​://doi.
supercomputing infrastructure of the NLHPC (ECM-02) org/10.1007/s0038​2-011-1187-x
Bozkurt D, Rondanelli R, Garreaud R, Arriagada A (2016) Impact of
warmer eastern tropical Pacific SST on the March 2015 Atacama
floods. Mon Weather Rev 144(11):4441–4460
References Bozkurt D, Rojas M, Boisier JP, Valdivieso J (2018a) Projected hydro-
climate changes over Andean basins in central Chile from down-
Alvarez-Garreton C, Menzoda PA, Boisier JPea (2018) The CAMELS- scaled CMIP5 models under the low and high emission scenarios.
CL dataset: catchment attributes and meteorology for large sample Clim Change 150(3–4):131–147. https​://doi.org/10.1007/s1058​
studies-Chile dataset. Hydrol Earth Syst Sci 22:5817–5846 4-018-2246-7
Bieniek PA, Bhatt US, Walsh JE, Rupp TS, Zhang J, Krieger JR, Vader Bozkurt D, Rondanelli R, Marín J, Garreaud R (2018b) Foehn event
R (2016) Dynamical downscaling of ERA-Interim temperature triggered by an atmospheric river underlies record-setting tem-
and precipitation for Alaska. J Appl Meteor Climatol 55:635–654. perature along continental Antarctica. J Geophys Res Atmos
https​://doi.org/10.1175/JAMC-D-15-0153.1 128(3):3871–3892. https​://doi.org/10.1002/2017J​D0277​96
Boisier JP, Rondanelli R, Garreaud R, Muñoz F (2016) Anthropogenic Carril A, Menéndez C, Remedio A et al (2012) Assessment of a multi-
and natural contributions to the Southeast Pacific precipitation RCM ensemble for South Eastern South America. Clim Dyn
39:2747–2768

13
 D. Bozkurt et al.

Comin AN, Schumacher V, Justino F, Fernrández A (2018) Impact Giorgetta MA et al (2013) Climate and carbon cycle changes from
of different microphysical parameterizations on extreme snowfall 1850 to 2100 in MPI-ESM simulations for the Coupled Model
events in the Southern Andes. Weather Clim Extrem. https​://doi. Intercomparison Project phase 5. J Adv Model Earth Syst
org/10.1016/j.wace.2018.07.001 5(3):572–597
Coppola E, Giorgi F, Raffaele F, Fuentes-Franco R, Giuliani G, LLo- Giorgi F, Marinucci MR, Bates GT (1993a) Development of a
part-Pereira M, Mamgain A, Mariotti L, Diro GT, Torma C (2014) second generation regional climate model (RegCM2). Part
Present and future climatologies in the phase i CREMA experi- I: Boundary layer and radiative transfer processes. Mon
ment. Clim Change 125(1):23–38. https:​ //doi.org/10.1016/j.atmos​ Weather Rev 121:2794–2813. https​: //doi.org/10.1175/1520-
env.2011.02.001 0493(1993)121<2794:DOASG​R>2.0.CO;2
Dee DP, Uppala SM, Berrisford P, Poli P, Kobayashi S, Andrae U, Giorgi F, Marinucci MR, Bates GT, DeCanio G (1993b) Develop-
Balmaseda MA, Balsamo G, Bauer P, Bechtold P, Beljaars ACM, ment of a second generation regional climate model (RegCM2).
van de Berg L, Bidlot J, Bormann N, Delsol C, Dragani R, Fuentes Part II: Convective processes and assimilation of lateral bound-
M, Geer AJ, Haimberger L, Healy SB, Hersbach H, Hólm EV, ary conditions. Mon Weather Rev 121:2814–2832. https​://doi.
Isaksen L, Káallberg P, Káhler M, Matricardi M, McNally AP, org/10.1175/1520-0493(1993)121<2814:DOASG​R>2.0.CO;2
Monge-Sanz BM, Morcrette JJ, Park BK, Peubey C, de Rosnay Giorgi F, Jones C, Asrar GR (2009) Addressing climate information
P, Tavolato C, Thépaut JN, Vitart F (2011) The ERA-Interim rea- needs at the regional level: the CORDEX framework. WMO
nalysis: configuration and performance of the data assimilation Bull 58:175–183
system. Q J R Meteorol Soc 137:553–597 Giorgi F, Coppola E, Solmon F, Mariotti L, Sylla MB, Bi X, Elguindi
Demaria EMC, Maurer EP, Thrasner B, Vicuñna S, Meza F (2013) N, Diro GT, Nair V, Giuliani G, Turuncoglu UU, Cozzini S,
Climate change impacts on an Alpine Watershed in Chile: do new Guttler I, O’Brien TA, Tawfik AB, Shalaby A, Zakey AS,
model projections change the story? J Hydrol 502:128–138. https​ Steiner AL, Stordal F, Sloan LC, Brankovic C (2012) RegCM4:
://doi.org/10.1016/j.jhydr​ol.2013.08.027 model description and preliminary tests over multiple CORDEX
DGA (2017) Actualización del balance hídrico nacional. Tech. Rep. domains. Clim Res 52:7–29. https​://doi.org/10.3354/cr010​18
SIT No. 417, Ministerio de Obras Públicas, Dirección General Giorgi F, Torma C, Coppola E, Ban N, Schar C, Somot S (2016)
de Aguas, División de Estudios y Planificación, Santiago, Chile. Enhanced summer convective rainfall at Alpine high elevations
Realizado por Universidad de Chile and Pontificia Universidad in response to climate warming. Nat Geosci 9:584–589. https​://
Católica de Chile doi.org/10.1038/ngeo2​761
Di Luca A, Elia R, Laprise R (2013) Potential for small scale added Grassi B, Redaelli G, Visconti G (2013) Arctic sea ice reduction and
value of RCMs downscaled climate change signal. Clim Dyn extreme climate events over the Mediterranean region. J Clim
40(3–4):601–618. https​://doi.org/10.1007/s0038​2-012-1415-z 26:10,101–10,110
Di Luca A, Argueso D, Evans JP, Laprise R (2016) Quantifying the Grell GA (1993) Prognostic evaluation of assumptions used by
overall added value of dynamical downscaling and the con- Cumulus parameterizations. Mon Weather Rev 121:764–787
tribution from different spatial scales. J Geophys Res Atmos Grell GA, Dudhia J, Stauffer DR (1994) Description of the fifth
121(4):1575–1590. https​://doi.org/10.1002/2015J​D0240​09 generation Penn State/NCAR Mesoscale Model (MM5). NCAR
Dickinson RE, Henderson-Sellers A, Kennedy PJ (1993) Biosphere- Tech. Note NCAR/TN-398+STR, NCAR, Boulder
atmosphere transfer scheme (BATS) Version 1e as coupled to the Guttler I, Stepanov I, Branković C, Nikulin G, Jones C (2015) Impact
NCAR Community Climate Model. NCAR Tech. Note NCAR/ of horizontal resolution on precipitation in complex orography
TN-387+STR, NCAR, Boulder simulated by the Regional Climate Model RCA3. Mon Weather
Falco M, Carril AF, Menéndez CG, Zaninelli PG, Li LZX (2018) Rev 143:3610–3627. https​://doi.org/10.1175/MWR-D-14-00302​
Assessment of CORDEX simulations over South America: added .1
value on seasonal climatology and resolution considerations. Clim Holtslag AAM, de Bruijn EIF, Pan HL (1990) A high resolution air
Dyn. https​://doi.org/10.1007/s0038​ mass transformation model for short-range weather forecasting.
Falvey M, Garreaud R (2009) Regional cooling in a warming world: Mon Weather Rev 118:1561–1575. https​://doi.org/10.1175/1520-
recent temperature trends in the Southeast Pacific and along the 0493(1990)118<1561:AHRAM​T>2.0.CO;2
west coast of subtropical South America (1979–2006). J Geophys Jones C, Giorgi F, Asrar GR (2011) The coordinated regional downs-
Res Atmos 114(D04):102 caling experiment: CORDEX, an international downscaling link
Fritsch JM, Chappell CF (1980) Numerical prediction of convectively to CMIP5. CLIVAR Exch 16:34–40
driven mesoscale pressure systems. I. Convective parameteriza- Jones PW (1999) First- and second-order conservative remapping
tion. J Geophys Res 37:1722–1733 schemes for grids in spherical coordinates. Mon Weather Rev
Gallardo L, Olivares G, Langner J, Aarhus B (2002) Coastal lows 127:2204–2210
and sulfur air pollution in Central Chile. Atmos Environ Karl TR, Nicholls N, Ghazi A (1999) CLIVAR/GCOS/WMO workshop
36(23):3829–3841 on indices and indicators for climate extremes: workshop sum-
Garreaud R (2009) The Andes climate and weather. Adv Geosci 7:1–9 mary. Clim Change 42:3–7
Garreaud R, Lopez P, Minvielle M, Rojas M (2010) Large-scale control Kato S, Loeb NG, Rose FG, Doelling DR, Rutan DA, Caldwell TE, Yu
on the Patagonian climate. Earth Planet Sci Lett 292:39–50. https​ L, Weller RA (2013) Surface irradiances consistent with ceres-
://doi.org/10.1016/j.epsl.2010.01.017 derived top-of-atmosphere shortwave and longwave irradiances. J
Garreaud R, Molina A, Farias M (2013) Andean uplift, ocean cooling Clim 26:2719–2740. https​://doi.org/10.1175/JCLI-D-12-00436​.1
and Atacama hyperaridity: a climate modeling perspective. J Clim Kiehl JT, Hack JJ, Bonan GB, Boville BA, Breigleb BP, Williamson
26(1):215–230. https​://doi.org/10.1175/JCLI-D-12-00001​.1 D, Rasch P (1996) Description of the NCAR Community Climate
Garreaud R, Falvey M, Montecinos A (2016) Orographic precipitation Model (CCM3). NCAR Tech. Note NCAR/TN-420+STR, NCAR,
in coastal southern Chile: mean distribution, temporal variability, Boulder
and linear contribution. J Hydrometeor 17:1185–1202 Lenaerts JTM, van den Broeke MR, van Wessem JM, van de Berg
Garreaud R, Alvarez-Garreton C, Barichivich J, Boisier JP, Duncan C, WJ, van Meijgaard E, van Ulft LH, Schaefer M (2014) Extreme
Galleguillos M, Zambrano-Bigiarini M (2017) The 2010–2015 precipitation and climate gradients in patagonia revealed by
mega drought in Central Chile: impacts on regional hydroclimate high-resolution regional atmospheric climate modeling. J Clim
and vegetation. Hydrol Earth Syst Sci 21:6307–6327 27:4607–4621. https​://doi.org/10.1175/JCLI-D-13-00579​.1

13
Dynamical downscaling over the complex terrain of southwest South America: present climate…

Marcella MP, Eltahir EA (2012) Modeling the summertime climate of Schmitz R (2005) Modelling of air pollution dispersion in Santiago de
Southwest Asia: the role of land surface processes in shaping the Chile. Atmos Environ 39(11):2035–2047
climate of semiarid regions. J Clim 25(2):704–719. https​://doi. Shi Y, Wang G, Gao X (2018) Role of resolution in regional climate
org/10.1175/2011J​CLI40​80.1 change projections over China. Clim Dyn 51(5–6):2375–2396.
Martens B, Miralles DG, Lievens H, van der Schalie R, de Jeu RAM, https​://doi.org/10.1007/s0038​
Fernandez-Prieto D, Beck HE, Dorigo WA, Verhoest NEC Silvestri G, Vera C (2009) Nonstationary impacts of the south-
(2017) Gleam v3: satellite-based land evaporation and root-zone ern annular mode on Southern Hemisphere climate. J Clim
soil moisture. Geosci Model Dev 10:1903–1925. https​://doi. 22(22):6142–6148
org/10.5194/gmd-2016-162 Solman SA (2013) Regional climate modeling over South
Mazzeo A, Huneeus N, Ordoñez C, Orfanoz-Cheuquelaf A, Menut America: a review. Adv Meteor 2013:1–13. https ​ : //doi.
L, Mailler S, Valari M, van der Gon HD, Gallardo L, Muñoz R, org/10.1155/2013/50435​7
Donoso R, Galleguillos M, Osses M, Tolvett S (2018) Impact of Solman SA, Sanchez E, Samuelsson P, da Rocha RP, Li L, Marengo
residential combustion and transport emissions on air pollution J, Pessacg NL, Remedio ARC, Chou SC, Berbery H, Le Treut
in santiago during winter. Atmos Environ 190:195–208. https​:// H, de Castro M, Jacob D (2013) Evaluation of an ensemble of
doi.org/10.1016/j.atmos​env.2018.06.043 regional climate model simulations over South America driven by
Middleton N (2003) Going to extremes. Pan Books, London the ERA-Interim reanalysis: model performance and uncertain-
Montesarchio M, Zollo AL, Bucchignani E, Mercogliano P, Castel- ties. Clim Dyn 41(5–6):1139–1157
lari S (2014) Performance evaluation of high-resolution regional Torma C, Giorgi F, Coppola E (2015) Added value of regional climate
climate simulations in the Alpine space and analysis of extreme modeling over areas characterized by complex terrain: precipita-
events. J Geophys Res Atmos 119:3222–3237. https ​: //doi. tion over the Alps. J Geophys Res Atmos 120(9):3957–3972. https​
org/10.1002/2013J​D0211​05 ://doi.org/10.1002/2014J​D0227​81
Onol B (2012) Effects of coastal topography on climate: high-resolu- Viale M, Garreaud R (2015) Orographic effects of the subtropical and
tion simulation with a regional climate model. Clim Res 52:159– extratropical Andes on upwind precipitating clouds. J Geophys
174. https​://doi.org/10.3354/cr010​77 Res Atmos 120:4962–4974. https​://doi.org/10.1002/2014J​D0230​
Pal JS, Small EE, Eltahir EAB (2000) Simulation of regional-scale 14
water and energy budgets: representation of surged cloud Viale M, Houze RA, Rasmussen KL (2013) Upstream orographic
and precipitation processes within RegCM. J Geophys Res enhancement of a narrow cold-frontal rainband approach-
105:29,579–29,594 ing the Andes. Mon Weather Rev 141:1708–1730. https​://doi.
Peterson TC, et al (2001) Report on the activities of the working group org/10.1175/MWR-D-12-00138​.1
on climate change detection and related rapporteurs 1998-2001. Vuille M, Franquist E, Garreaud R, Casimiro WSL, Cáceres B (2015)
WMO Tech. Report Rep. WCDMP-47, WMO-TD 1071, WMO, Impact of the global warming hiatus on Andean temperature. J
Geneve, Switzerland Geophys Res Atmos 120(9):3745–3757
Rojas M (2006) Multiply nested regional climate simulation for south- Walker MD, Diffenbaugh NS (2009) Evaluation of high-resolution
ern South America: sensitivity to model resolution. Mon Weather simulations of daily-scale temperature and precipitation over the
Rev 134:2208–2223. https​://doi.org/10.1175/MWR31​67.1 United States. Clim Dyn 33:1131–1147
Rutllant J, Fuenzalida H (1991) Synoptic aspects of the central Chile Zeng X, Zhao M, Dickinson RE (1998) Intercomparison of bulk aero-
rainfall variability associated with the Southern Oscillation. Int dynamic algorithms for the computation of sea surface fluxes
J Climatol 11:63–76 using TOGA COARE and TAO data. J Clim 11:2628–2644.
Rutllant J, Fuenzalida H, Aceituno P (2003) Climate dynamics along https ​ : //doi.org/10.1175/1520-0442(1998)011<2628:IOBAA​
the arid northern coast of Chile: The 1997–1998 Diclima Experi- F>2.0.CO;2
ment. J Geophys Res 108:4538–4542
Saide PE, Carmichael GR, Spak SN, Gallardo L, Osses AE, Mena- Publisher’s Note Springer Nature remains neutral with regard to
Carrasco MA, Pagowski M (2011) Forecasting urban PM10 jurisdictional claims in published maps and institutional affiliations.
and PM2.5 pollution episodes in very stable nocturnal condi-
tions and complex terrain using WRF-Chem CO tracer model.
Atmos Environ 45:2769–2780. https​://doi.org/10.1016/j.atmos​
env.2011.02.001

13