15 MAY 2015 MANZANAS ET AL.

4171

Statistical Downscaling in the Tropics Can Be Sensitive to Reanalysis

Choice: A Case Study for Precipitation in the Philippines

R. MANZANAS
Grupo de Meteorologı́a, Dpto. Matemática Aplicada y CC. Computación, Universidad de Cantabria, Santander, Spain

S. BRANDS
Grupo de Meteorologı́a, Instituto de Fı́sica de Cantabria, CSIC–Universidad de Cantabria, Santander, Spain

D. SAN-MARTÍN
Predictia Intelligent Data Solutions, Santander, Spain

A. LUCERO AND C. LIMBO

Philippine Atmospheric, Geophysical and Astronomical Services Administration, Quezon City, Philippines

J. M. GUTIÉRREZ
Grupo de Meteorologı́a, Instituto de Fı́sica de Cantabria, CSIC–Universidad de Cantabria, Santander, Spain

(Manuscript received 25 April 2014, in final form 19 January 2015)

ABSTRACT

This work shows that local-scale climate projections obtained by means of statistical downscaling are sensitive to
the choice of reanalysis used for calibration. To this aim, a generalized linear model (GLM) approach is applied to
downscale daily precipitation in the Philippines. First, the GLMs are trained and tested separately with two distinct
reanalyses (ERA-Interim and JRA-25) using a cross-validation scheme over the period 1981–2000. When the
observed and downscaled time series are compared, the attained performance is found to be sensitive to the
reanalysis considered if climate change signal–bearing variables (temperature and/or specific humidity) are in-
cluded in the predictor field. Moreover, performance differences are shown to be in correspondence with the
disagreement found between the raw predictors from the two reanalyses. Second, the regression coefficients
calibrated either with ERA-Interim or JRA-25 are subsequently applied to the output of a global climate model
(MPI-ECHAM5) in order to assess the sensitivity of local-scale climate change projections (up to 2100) to re-
analysis choice. In this case, the differences detected in present climate conditions are considerably amplified,
leading to ‘‘delta-change’’ estimates differing by up to 35% (on average for the entire country) depending on the
reanalysis used for calibration. Therefore, reanalysis choice is an important contributor to the uncertainty of local-
scale climate change projections and, consequently, should be treated with as much care as other better-known
sources of uncertainty (e.g., the choice of the GCM and/or downscaling method). Implications of the results for the
entire tropics, as well as for the model output statistics downscaling approach are also briefly discussed.

1. Introduction climate information required by impact and adaptation

studies. These techniques, however, have been de-
Statistical downscaling (SD) techniques are nowadays
veloped and applied almost exclusively for extratropical
routinely applied to translate coarse-resolution output
regions (Hewitson and Crane 1996; Wilby and Wigley
from global climate models (GCMs) to local-scale
1997; Trigo and Palutikof 2001; Hanssen-Bauer et al.
2005; Fowler et al. 2007; Maraun et al. 2010; Gutiérrez
et al. 2013). Nevertheless, for low-latitude regions
Corresponding author address: R. Manzanas, Grupo de Meteorología,
Dpto. Matemática Aplicada y Ciencias de la Computación, Universidad (e.g., tropical Africa or southeast Asia), where the
de Cantabria, Avda. los Castros, s/n, 39005, Santander, Spain. demand for reliable local-scale climate information is
E-mail: rmanzanas@ifca.unican.es of paramount importance due to a large vulnerability

DOI: 10.1175/JCLI-D-14-00331.1

Ó 2015 American Meteorological Society

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC
4172 JOURNAL OF CLIMATE VOLUME 28

to changing environmental factors (Wilby et al. 2009), The paper is outlined as follows: In section 2, the con-
studies are rare or even nonexistent to date since sidered datasets are described and a brief introduction to
manifold problems still hinder the successful applica- the precipitation climate of the Philippines is provided.
tion of SD in these regions (Hewitson et al. 2014). The applied downscaling technique is described in
In the so-called perfect-prog approach, local-scale section 3 and the results are presented through section 4.
climate variability, typically represented by (gridded) Section 5 provides the conclusions and a brief discussion
weather station records, is statistically linked to quasi- on the implications for the entire tropics as well as the
observations from reanalysis datasets (Marzban et al. model output statistics downscaling approach.
2006). The success of perfect-prog schemes in the
extratropics relies on the fact that a large fraction of
local-scale climate variability can be described by at- 2. Data
mospheric phenomena operating on spatial scales on
a. Predictands
the order of thousands of kilometers, typically having
a lifetime of several days. At this scale, reanalysis datasets Daily precipitation amounts from 42 gauges main-
are known to be skillful, in the sense that their spatio- tained by the Philippine Atmospheric, Geophysical and
temporal resolution captures the relevant processes such Astronomical Services Administration (PAGASA)
as extratropical cyclones and the associated fronts were considered as predictand data for the period 1981–
(Grotch and MacCracken 1991; Widmann et al. 2003). At 2000 (see Fig. 1b). These station time series, which in the
lower latitudes, however, the atmospheric drivers of local- following are classified into the four precipitation cli-
scale climate variability operate on much finer scales mate types (CTs) defined in Coronas (1920), were se-
(both spatial and temporally) and are generally poorly lected after a rigorous quality control, thus minimizing
captured by reanalysis datasets (Manzanas et al. 2014). the predictand-induced uncertainty (Hewitson et al.
Moreover, observational coverage is generally sparse in 2014). As can be seen in Fig. 1c, precipitation along the
the tropics, leading to considerable differences between coastlines of the northern part of the archipelago (CT1
distinct reanalyses (Trenberth et al. 2001; Sterl 2004; and CT2) exhibits a strong seasonal cycle, which is driven
Brands et al. 2012, 2013) and to errors with respect to by alternating monsoonal winds. In particular, during the
observational records (Manzanas et al. 2014), which in southwest monsoon (June–September), precipitation
turn can complicate the detection of a relationship with peaks at the stations pertaining to CT1 while CT2 is af-
the local-scale climate variability. fected by relative dryness. However, the opposite is the
Therefore, the present work tests whether reanalysis case during the northeast monsoon (October–February).
choice is relevant for the application of SD in the tropics. During the dry months (March–May), easterly winds
A generalized linear model (GLM) approach is prevail, leading to orographic precipitation along the
separately calibrated for two distinct reanalyses (ERA- mountain ranges in the east of the archipelago (see
Interim and JRA-25) in order to downscale daily Fig. 1a) and to relatively high precipitation amounts for
precipitation over the Philippines, using a long-term, the stations pertaining to CT2. At the stations pertaining
quality-controlled precipitation dataset that essentially to CT3 and CT4 (mainly situated in the center and south
eliminates predictand-induced uncertainty (Hewitson of the archipelago), precipitation is bounded to mesoscale
et al. 2014). Because of its geographical location between dynamics and is not directly driven by the monsoons,
the monsoonal and inner tropics, the Philippines pro- leading to a weak seasonal cycle. Additionally, in-
vides an ideal testbed for SD studies. terannual variability is larger for CT1 and CT2 than for
First, following a cross-validation scheme for the period CT3 and CT4 (Fig. 1d). For a comprehensive description
1981–2000, the downscaling results are shown to be sen- of the climate in the Philippines, the interested reader is
sitive to reanalysis choice if climate change signal–bearing referred to Coronas (1920), Flores and Balagot (1969),
variables such as temperature and/or specific humidity are and Kintanar (1984) as well as to the PAGASA website
used as predictors. Second, when the reanalyses- (http://www.pagasa.dost.gov.ph/).
calibrated coefficients are applied to predictor data
b. Predictors
from a GCM (MPI-ECHAM5)—in which case signal-
bearing predictor variables should be applied in order to Atmospheric variables describing circulation, mois-
capture the ‘‘correct’’ climate change signal (Goodess and ture, and convection are generally considered to be
Palutikof 1998; Wilby et al. 1998)—the sensitivity to re- among the most informative predictors for perfect-prog
analysis choice is largely amplified, leading to differences SD of precipitation (Charles et al. 1999; Timbal et al.
in the projected ‘‘deltas’’ of up to 35% (on average for the 2003; Bürger and Chen 2005; Cavazos and Hewitson
entire country) for both reanalyses. 2005; Dibike and Coulibaly 2005; Haylock et al. 2006;

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

15 MAY 2015 MANZANAS ET AL. 4173

FIG. 1. (a) Topography of the Philippines. (b) Location of the 42 gauges considered, classified into the four pre-
cipitation climate types (CTs) described in the text. Each CT is indicated by a specific color. (c) Intra- and (d) in-
terannual variability of spatial average precipitation amount for each CT (period: 1981–2000).

Hewitson and Crane 2006; Fowler et al. 2007; Hertig and included into the predictor field in order to 1) improve
Jacobeit 2008; Timbal and Jones 2008; Sauter and the statistical link-function (i.e., the predictive potential
Venema 2011). If SD is applied in climate change con- of the SD method) and 2) capture the ‘‘correct’’ climate
ditions (i.e., to predictor data obtained from a given change signal (Goodess and Palutikof 1998; Wilby et al.
GCM), the GCM is assumed to perfectly reproduce 1998).
the same climatological properties provided by the re- With these precepts in mind, and after consulting the
analysis used for calibration (Hewitson and Crane 1996; expertise of local meteorologists as well as the results
Wilby et al. 2004). In other words, the ‘‘performance’’ of from previously published studies (Kang et al. 2007;
the GCM (Giorgi and Mearns 2002) must be evaluated Chu et al. 2008; Paul et al. 2008; Chu and Yu 2010), a set
for the relevant predictor variables. An important di- of different predictor combinations was chosen (see
lemma of perfect-prog SD is that GCMs generally per- Table 1). These combinations consist of circulation var-
form better for circulation and temperature variables iables alone (zonal wind component at 850 and 300 hPa;
than for moisture ones (Räisänen 2007; Brands et al. P1: U850, U300), circulation and specific humidity (P2:
2011, 2013). Yet, moisture information should be U850, U300, Q850), circulation and temperature (P3:

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

4174 JOURNAL OF CLIMATE VOLUME 28

TABLE 1. Considered predictor combinations. known as logistic regression) is used to downscale daily
precipitation occurrence (a threshold of 0.5 mm was
Abbreviation Predictor variables
used to define occurrence). Probabilities equal or
P1 U850, U300
greater (smaller) than 0.5 are considered as rainfall oc-
P2 U850, U300, Q850
P3 U850, U300, T850 currences (absences). Second, a GLM with gamma error
P4 U850, U300, Q850, T850 distribution and log link-function is applied to down-
Q850 Q850 scale daily precipitation amount. Unlike in other studies,
T850 T850 the stochastic component of the GLM is excluded from
each of the two models (occurrence and amount); that is,
expected values are predicted in any case. This is done to
U850, U300, T850) and circulation, specific humidity, and isolate the effect of reanalysis uncertainty on the
temperature (P4: U850, U300, Q850, T850). In addition, downscaling results.
Q850 and T850 were considered as single predictor var- For each gauge, predictor data at the four nearest grid
iables. Note that Q850 is used instead of column in- points are considered for both the occurrence and
tegrated water vapor or precipitable water since the latter amount models. For the case of the reanalyses and the
variables are usually not provided by the common GCM GCM in the control period, each predictor variable is
databases. standardized grid box by grid box to have zero mean and
The predictor variables listed in Table 1 were ob- unit variance. Standardization brings the first- and
tained from two distinct reanalyses and one GCM: The second-order moments of the reanalysis and GCM data
European Centre for Medium-Range Weather Fore- into agreement and thereby provides a better approxi-
casts (ECMWF) ERA-Interim reanalysis (Dee et al. mation for the assumption of ‘‘perfect’’ GCM perfor-
2011), the Japanese 25-year Reanalysis (JRA-25) mance than using untransformed data. The GCM
(Onogi et al. 2007), and the Max Planck Institute (MPI) scenario data are standardized by removing the mean of
ECHAM5 GCM (Giorgetta et al. 2006); see the ac- the control period from the mean of the corresponding
knowledgments for data sources. For the case of scenario period and dividing by the standard deviation
ECHAM5, control and A1B scenario data from the of the control period.
third transient run developed within the ENSEMBLES To avoid overfitting, a k-fold cross-validation ap-
project were retrieved. To keep consistency between the proach (Gutiérrez et al. 2013) was followed, with k 5 4
time steps available for both reanalyses and the GCM, nonoverlapping test periods of five years each, covering
daily instantaneous values at 0000 UTC were chosen in the full period 1981–2000. To circumvent spurious trend
all cases. Because of distinct native resolutions, the effects, the five years forming each test period were
predictor data from all sources were regridded onto randomly chosen.
a common regular 28 grid using bilinear interpolation.

4. Results
3. Downscaling technique
a. Reanalysis differences in the predictor data
The downscaling technique used here to build transfer
functions from the predictors (x1 , . . . , xn ) to the pre- The top panel of Fig. 2 shows a comparison between
dictand (y) is based on generalized linear models ERA-Interim (taken as reference) and JRA-25 for
(Nelder and Wedderburn 1972), which allow for non- the four predictor variables in Table 1 over the Co-
normal error distributions. The conditional expected ordinated Regional Climate Downscaling Experiment
value of the predictand given the predictors is linked via (CORDEX)-East Asian domain (http://wcrp-cordex.ipsl.
a monotonic function to a linear combination of the jussieu.fr/images/pdf/cordex_regions.pdf) for the period
predictors b0 1 b1 x1 1


1 bp xp , where b0 , . . . , bp 1981–2000. The left column shows the mean difference
are the regression coefficients. These models have been (bias) between both reanalyses, expressed as a percent-
used in numerous previous downscaling studies dealing age of ERA-Interim’s standard deviation. The middle
with precipitation (e.g., Brandsma and Buishand 1997; column displays the ratio of variances (RV), defined as
Chandler and Wheater 2002; Abaurrea and Asín 2005; s2J /s2E , where s2J (s2E ) is the variance of JRA-25 (ERA-
Fealy and Sweeney 2007; Hertig et al. 2013). Interim), respectively. In the right column, the Pearson
In this work, the two-stage implementation commonly correlation coefficient (r) between the two reanalyses is
used for precipitation downscaling is applied (see, e.g., depicted.
Chandler and Wheater 2002). First, a GLM with Ber- As can be seen, there are appreciable differences
noulli error distribution and logit link-function (also (systematically lower for U850 and U300 than for Q850

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

15 MAY 2015 MANZANAS ET AL. 4175

FIG. 2. (top) Comparison between ERA-Interim (taken as reference) and JRA-25 for the four predictor variables in Table 1 (in rows)
over the period 1981–2000. An explanation of the applied comparison metrics (in columns) is provided in the text. The Philippines
archipelago is indicated by the black boxes. (bottom right) Pearson correlation coefficient (r) between the two reanalysis time series, as
a function of latitude (displayed are zonal averages) for the Philippines archipelago. Different colors indicate different predictors. (bottom
left) Grid box coordinates used for computing the zonal averages.

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

4176 JOURNAL OF CLIMATE VOLUME 28

FIG. 3. Cross-validation results for each CT as measured by the Spearman correlation coefficient (rs) between
downscaled and observed daily precipitation amount (period: 1981–2000). Different colors correspond to different
predictor combinations (see the legend) and solid (dashed) lines refer to the results obtained from using ERA-
Interim (JRA-25) predictor data. The mean and standard deviation of the four cross-validation results are indicated
by lines and error bars respectively. For each CT, results are sorted by the latitude of the stations (decreasing from left
to right). CT-specific spatial average values are shown on the right-hand side of each panel; points (asterisks) cor-
respond to the results from using ERA-Interim (JRA-25) predictor data.

and T850) between both reanalyses for the three vali- Philippines archipelago. The gridbox coordinates are
dation measures considered, indicating that the perfect- mapped on the left-hand side and r as a function of
prog assumption (reanalysis data reflecting ‘‘real’’ latitude is displayed on the right-hand side. Noticeably,
large-scale atmospheric conditions) does not hold for U850 exhibits values around 0.95 at all latitudes, which
the area under study. Nevertheless, with respect to their indicates that both reanalyses are in nearly perfect
application for SD, recall that the reanalysis time series agreement for this variable. However, a north–south
are standardized to have zero mean and unit variance gradient is found for the remaining variables. In partic-
before ‘‘entering’’ the downscaling scheme (section 3). ular, correlations for T850 and Q850 drop from 0.95 to
Consequently, differences in the mean and variance be- 0.70 and from 0.75 to 0.50, respectively, probably re-
tween the two reanalyses (left and middle columns) do flecting the increasing influence of subgrid processes—
not affect the SD results, whereas differences in the third- subject to reanalysis/model-dependent parameterization
and fourth-order moments (i.e., skewness and kurtosis; schemes—toward the equator.
see, e.g., Brands et al. 2011) and in day-to-day variations
b. Differences in cross-validation results
(right column) remain and are expected to affect them.
In the bottom panel of Fig. 2, the zonally averaged r Figure 3 displays the Spearman correlation coefficient
between the predictor time series from JRA-25 and (rs) between daily observed and downscaled pre-
ERA-Interim is displayed for the specific case of the cipitation time series over the period 1981–2010 for

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

15 MAY 2015 MANZANAS ET AL. 4177

different predictor combinations—P1 (U850, U300), of the archipelago, whereas a gradual decrease is ob-
Q850, T850, and P4 (U850, U300, Q850, T850)—when served toward the south. For circulation predictors only,
considering predictor data from ERA-Interim and JRA- both reanalyses perform similarly (first row). However,
25 (solid and dashed lines, respectively). In each panel, if Q850 (T850) is added to circulation, better results are
the results for a specific CT are shown. Lines and error obtained for ERA-Interim (JRA-25) (second and third
bars correspond to the mean and standard deviation of rows, respectively). Notably, for the case of including
the cross-validation results (computed upon the four T850, the advantage of JRA-25 over ERA-Interim is
folds considered). Along the x axis, stations are sorted most obvious along the eastern coastline. When consid-
by decreasing latitude (from left to right). On the right- ering the ‘‘full’’ predictor combination (P4), ERA-
hand side, the CT-averaged rs are indicated. Points Interim systematically outperforms JRA-25 at all stations.
(asterisks) correspond to ERA-Interim (JRA-25) pre- To further assess the increase in predictive potential
dictor data. from adding temperature and moisture information to
For both reanalyses, the combination of circulation, circulation, Fig. 5 shows the difference in rs [d(rs)] ob-
humidity, and temperature predictors (P4) yields high- tained when adding Q850 and T850 separately (P2 and
est correlation coefficients. The predictive potential is P3, respectively) and in combination (P3) to the ‘‘basic’’
slightly lower for using circulation variables alone (P1) circulation variables (P1). Results for calibrating with
and further decreases if circulation is excluded from the ERA-Interim and JRA-25 are given in the left and
predictor field, that is, for using Q850 and T850 sepa- middle column, respectively. Additionally, the corre-
rately or in combination (the latter not shown). sponding differences—JRA-25 minus ERA-Interim—
Moreover, for the sole use of circulation variables are shown in the right column. Positive (negative) values
(P1), the downscaling results are generally not sensitive indicate a larger increment for JRA-25 (ERA-Interim).
to reanalysis choice, except for the stations situated in In congruence with Figs. 3 and 4, the performance
the south (CT4). This is in agreement with the small improvement attained when adding Q850 (T850) is
differences found between ERA-Interim and JRA-25 larger for ERA-Interim (JRA-25) than for JRA-25
for U300 and U850 as well as with the slight north–south (ERA-Interim). Moreover, when including Q850 1
uncertainty gradient detected for U300 (see Fig. 2). T850, the improvement is larger for ERA-Interim than
However, for Q850 and T850, appreciable reanalysis- for JRA-25. The previous results prove that, depending
induced differences are observed. In particular, Q850 on the choice of reanalysis, up to 0.10 correlation points
from ERA-Interim yields better results than Q850 from can be missed on the local scale for particular predictor
JRA-25, whereas the opposite is the case for T850 combinations.
(with the exception of CT1). This indicates that the
c. Differences in climate change projections
‘‘real’’ statistical relationship between Q850 (T850) and
local-scale precipitation is more accurately captured by In this section it will be shown that local-scale climate
ERA-Interim (JRA-25). Moreover, when considering the projections obtained by SD are sensitive to the choice of
‘‘best’’ predictor combination (P4), results are systemat- reanalysis used for calibration. To this aim, the regression
ically better for ERA-Interim than for JRA-25. Notably, coefficients obtained from separately calibrating the
the southward loss of predictive potential occurring in all GLMs with either ERA-Interim or JRA-25 are applied to
CTs except CT2 is in agreement with the southward in- predictor data from MPI-ECHAM5. This is done for the
crease of reanalysis uncertainty (Figs. 2 and 3). reference period 1981–2000 (using control run data)
For a geographical overview of these results, Fig. 4 shows and for three different future periods (2011–40, 2041–70,
the mean pointwise cross-validation rs when considering and 2071–2100), using scenario run data (A1B, run 3).
ERA-Interim (left column) and JRA-25 (middle column) The underlying assumption of this procedure is that the
predictor data, with each row corresponding to a specific predictor–predictand relationships obtained above remain
predictor combination. The corresponding differences— stationary in time (Vrac et al. 2007).
JRA-25 minus ERA-Interim—are displayed in the right The climate change projections are obtained by means
column, so positive (negative) values indicate that JRA-25 of the delta method, that is, by subtracting the reference/
(ERA-Interim) is more appropriate for SD. Because of the control period’s mean from the mean of the corre-
lower predictive potential described above, results for the sponding target scenario period (Räisänen 2007). Deltas
single predictor variables (Q850 and T850) are not in- are shown as relative (%) deviations from the mean in
cluded in Fig. 4. the reference period (0% 5 no deviation).
The spatial pattern of predictive potential is similar Figure 6 shows, from left to right, three panels, one for
for the four predictor combinations. Highest rs values each of the future periods considered. In each panel, the
are obtained in the north and along the eastern coastline deltas projected by applying the coefficients learned from

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

4178 JOURNAL OF CLIMATE VOLUME 28

FIG. 4. Spearman correlation coefficient between observed and downscaled daily

precipitation amount for different predictor combinations (in rows), when consid-
ering predictor data from (left) ERA-Interim and (middle) JRA-25. Displayed is the
mean value of the four cross-validation results. Black frames indicate the better-
performing reanalysis for a specific predictor combination. (right) JRA-25’s
performance minus ERA-Interim’s performance. The numbers in each panel cor-
respond to the spatial average values for all stations or those stations pertaining to
a specific CT (All and CT1–4, respectively).

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

15 MAY 2015 MANZANAS ET AL. 4179

FIG. 5. Performance improvement (with respect to P1) for different predictor combinations (in rows), when
considering predictor data from (left) ERA-Interim and (middle) JRA-25. For each row, the black frame in-
dicates the better-performing reanalysis. (right) The performance improvement differences (JRA-25 minus
ERA-Interim). The numbers in each panel correspond to the spatial average values for all stations or those
stations pertaining to a specific CT (All and CT1–4, respectively).

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

4180 JOURNAL OF CLIMATE VOLUME 28

FIG. 6. The effect of reanalysis uncertainty on future precipitation projections. Displayed are precipitation deltas for three future
periods (from left to right, 2011–40, 2041–70, and 2071–2100, all with respect to the control period 1981–2000). Each future period makes
up a column. For each column, the left (middle) panel shows the deltas obtained from applying the regression coefficients learned from
ERA-Interim (JRA-25) to predictor data from MPI-ECHAM5. The right column displays the difference between the JRA-25’s delta and
ERA-Interim’s delta. The numbers in each panel correspond to the spatial average values for all stations or those stations pertaining to
a specific CT (All and CT1–4, respectively).

ERA-Interim (JRA-25) are shown in the left (middle) by MPI-ECHAM5) over the target region is not sensi-
column, while the corresponding differences (JRA-25’s tive to climate change.
delta minus ERA-Interim’s delta) are provided in the However, if Q850 and/or T850 are added to circula-
right column; each row corresponds to a particular pre- tion, the projected deltas increase as a function of lead
dictor combination. The numbers in each map indicate time (i.e., are larger for the end of the century; see the
the spatial mean value for all stations (All) or those sta- second, third, and fourth rows). This holds valid for us-
tions pertaining to a specific CT (CT1–CT4, respectively). ing Q850 and T850 as separate predictors (not shown).
As can be seen, a negligible delta is found for any Remarkably, precipitation deltas for Q850 and T850 are
future period if precipitation is downscaled from circu- larger than for P4, indicating that the inclusion of cir-
lation variables alone (first row). Note that this is in culation damps the climate change signal (not shown).
agreement with the time evolution of U850 and The fact that Q850—either alone (not shown) or in
U300, which is virtually constant throughout the whole combination with U850 and U300 (P2 in Fig. 6)—leads
twenty-first century (first and second rows in Fig. 7), to the largest delta differences proves that the down-
indicating that the large-scale circulation (as simulated scaling results are especially sensitive to reanalysis

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

15 MAY 2015 MANZANAS ET AL. 4181

FIG. 7. Mean value—as simulated by MPI-ECHAM5 (A1B scenario, run 3)—in the three considered future periods
(2011–40, 2041–70, and 2071–2100, in columns) for each of the predictor variables listed in Table 1 (in rows).

choice when this variable is included in the predictor change signal imposed by the GCM (cf. Figs. 6 and 7).
field. For P2, reanalysis-induced delta differences reach Also, the magnitude of the projected deltas seems to be
35% (45%) for the entire country (CT1) at the end of related to the cross-validation results of section 4b. In
the century (2071–2100). particular, larger deltas are obtained for the ‘‘better’’
Finally, note that the reanalysis-induced differences in performing reanalysis, that is, ERA-Interim (JRA-25)
the downscaled time series are proportional to the climate when Q850 (T850) is added to circulation.

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

4182 JOURNAL OF CLIMATE VOLUME 28

5. Conclusions and discussion for the model output statistics approach, in which GCMs
are nudged to reanalysis data in order to force them to
In this study, a generalized linear model (GLM) ap- follow the ‘‘observed’’ large-scale variability (Eden et al.
proach is applied to downscale daily precipitation in the 2012). Here, it has been shown that the ‘‘real’’ large-
Philippines. To explore the effect of reanalysis un- scale atmospheric variability in the tropics is likely to be
certainty on statistical downscaling (SD), two distinct misrepresented by reanalyses and, consequently, also by
reanalysis datasets are used to obtain the link functions the aforementioned nudged GCMs. Finally, since re-
(regression coefficients) relating the local-scale pre- gional climate models can be nested into different re-
dictands to the large-scale predictors. When comparing analysis datasets, reanalysis uncertainty is also likely to
observed and downscaled daily precipitation time series affect the dynamical downscaling approach (Park et al.
over the period 1981–2000 using a cross-validation scheme, 2013).
results are found to be sensitive to the reanalysis dataset
selected for calibration, which is in agreement with the Acknowledgments. The authors are grateful to the
few previous studies addressing this issue (Koukidis and free distribution of the ECMWF ERA-Interim (http://www.
Berg 2009; Hofer et al. 2012; Park et al. 2013). However, ecmwf.int/en/research/climate-reanalysis/era-interim), JMA
with spatial average (local scale) correlation differences JRA-25 (http://jra.kishou.go.jp/JRA-25/index_en.html), and
of 0.03 (0.10) at the utmost, this sensitivity is relatively MPI-ECHAM5 data (http://cera-www.dkrz.de/WDCC/
small at this point. ui/Compact.jsp?acronym5ENSEMBLES_MPEH5_
The reanalysis-calibrated coefficients are subsequently SRA1B_3_D) and acknowledge PAGASA for the ob-
applied to predictor data from a global climate model servational data provided. This study was supported by
(GCM) in order to generate local-scale climate pro- the EU projects QWeCI and SPECS, funded by the
jections for the whole twenty-first century. In this case, European Commission through the Seventh Framework
the reanalysis-induced differences detected in present Programme for Research under Grant Agreements
climate conditions are considerably amplified when 243964 and 308378, respectively. RM also acknowledges
signal-bearing variables—which are indispensable for the EU project EUPORIAS, funded by the European
capturing the correct climate change signal—are included Commission through the Seventh Framework Pro-
in the predictor field. In particular, the projected deltas gramme for Research under Grant Agreement 308291.
for the end of the century (2071–2100 minus 1981–2000) SB is grateful to the CSIC-JAE-Predoc Program for
are found to differ by up to 35% (on average for the financial support.
whole country) for the two reanalyses considered.
Therefore, the choice of reanalysis used for calibration in
REFERENCES
perfect-prog SD is an important contributor to the un-
certainty of local-scale climate change projections and, Abaurrea, J., and J. Asín, 2005: Forecasting local daily pre-
consequently, should be treated with as much care as cipitation patterns in a climate change scenario. Climate Res.,
other well-known uncertainty sources, such as the choice 28, 183–197, doi:10.3354/cr028183.
Brands, S., S. Herrera, D. San-Martín, and J. M. Gutiérrez, 2011:
of GCM or downscaling method (Dibike and Coulibaly Validation of the ENSEMBLES global climate models over
2005; Chen et al. 2012). southwestern Europe using probability density functions,
Although these conclusions have been deduced for from a downscaling perspective. Climate Res., 48, 145–161,
a specific region (the Philippines), they are likely to hold doi:10.3354/cr00995.
valid for the entire tropics since previous studies point ——, J. M. Gutiérrez, S. Herrera, and A. S. Cofiño, 2012: On the
use of reanalysis data for downscaling. J. Climate, 25, 2517–
out that reanalysis uncertainty is a general problem at 2526, doi:10.1175/JCLI-D-11-00251.1.
low latitudes, especially for climate change signal– ——, S. Herrera, J. Fernández, and J. M. Gutiérrez, 2013: How well
bearing predictor variables on daily time scale (Brands do CMIP5 Earth system models simulate present climate
et al. 2012, 2013). Also, the presented results rely on conditions in Europe and Africa? Climate Dyn., 41, 803–817,
a single GCM and therefore should be reconfirmed with doi:10.1007/s00382-013-1742-8.
Brandsma, T., and T. A. Buishand, 1997: Statistical linkage of
alternative GCMs (with distinct model physics) in future daily precipitation in Switzerland to atmospheric circula-
studies. Besides, an exhaustive assessment on the pre- tion and temperature. J. Hydrol., 198, 98–123, doi:10.1016/
dictive potential of alternative predictor variables in the S0022-1694(96)03326-4.
tropics, such as velocity potential or streamfunction, and Bürger, G., and Y. Chen, 2005: Regression-based downscaling of
on the corresponding differences induced by reanalysis spatial variability for hydrologic applications. J. Hydrol., 311,
299–317, doi:10.1016/j.jhydrol.2005.01.025.
choice might be a useful future task. Cavazos, T., and B. C. Hewitson, 2005: Performance of NCEP–
Apart from being relevant for perfect-prog SD, NCAR reanalysis variables in statistical downscaling of daily
reanalysis uncertainty is expected to be equally relevant precipitation. Climate Res., 28, 95–107, doi:10.3354/cr028095.

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

15 MAY 2015 MANZANAS ET AL. 4183

Chandler, R. E., and H. S. Wheater, 2002: Analysis of rainfall for their robust application under climate change conditions.
variability using generalized linear models: A case study from J. Climate, 26, 171–188, doi:10.1175/JCLI-D-11-00687.1.
the west of Ireland. Water Resour. Res., 38, 1192, doi:10.1029/ Hanssen-Bauer, I., C. Achberger, R. E. Benestad, D. Chen, and
2001WR000906. E. J. Forland, 2005: Statistical downscaling of climate scenar-
Charles, S. P., B. C. Bates, P. H. Whetton, and J. P. Hughes, 1999: ios over Scandinavia. Climate Res., 29, 255–268, doi:10.3354/
Validation of downscaling models for changed climate con- cr029255.
ditions: Case study of southwestern Australia. Climate Res., Haylock, M. R., G. C. Cawley, C. Harpham, R. L. Wilby, and C. M.
12, 1–14, doi:10.3354/cr012001. Goodess, 2006: Downscaling heavy precipitation over the
Chen, H., C. Y. Xu, and S. Guo, 2012: Comparison and evaluation United Kingdom: A comparison of dynamical and statistical
of multiple GCMs, statistical downscaling and hydrological methods and their future scenarios. Int. J. Climatol., 26, 1397–
models in the study of climate change impacts on runoff. 1415, doi:10.1002/joc.1318.
J. Hydrol., 434, 36–45, doi:10.1016/j.jhydrol.2012.02.040. Hertig, E., and J. Jacobeit, 2008: Assessments of Mediterranean
Chu, J.-L., and P.-S. Yu, 2010: A study of the impact of climate precipitation changes for the 21st century using statistical
change on local precipitation using statistical downscaling. downscaling techniques. Int. J. Climatol., 28, 1025–1045,
J. Geophys. Res., 115, D10105, doi:10.1029/2009JD012357. doi:10.1002/joc.1597.
——, H. Kang, C.-Y. Tam, C.-K. Park, and C.-T. Chen, 2008: ——, S. Seubert, A. Paxian, G. Vogt, H. Paeth, and J. Jacobeit,
Seasonal forecast for local precipitation over northern Taiwan 2013: Changes of total versus extreme precipitation and dry
using statistical downscaling. J. Geophys. Res., 113, D12118, periods until the end of the twenty-first century: Statistical
doi:10.1029/2007JD009424. assessments for the Mediterranean area. Theor. Appl. Cli-
Coronas, J., 1920: The climate and weather of the Philippines, matol., 111, 1–20, doi:10.1007/s00704-012-0639-5.
1903–1918. Census of the Philippine Islands: 1918, Bureau of Hewitson, B. C., and R. G. Crane, 1996: Climate downscaling:
Printing, Manila, 291–467. Techniques and application. Climate Res., 7, 85–95, doi:10.3354/
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: cr007085.
Configuration and performance of the data assimilation sys- ——, and ——, 2006: Consensus between GCM climate change
tem. Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/ projections with empirical downscaling: Precipitation down-
qj.828. scaling over South Africa. Int. J. Climatol., 26, 1315–1337,
Dibike, Y. B., and P. Coulibaly, 2005: Hydrologic impact of climate doi:10.1002/joc.1314.
change in the Saguenay watershed: Comparison of down- ——, J. Daron, R. G. Crane, M. F. Zermoglio, and C. Jack, 2014:
scaling methods and hydrologic models. J. Hydrol., 307, 145– Interrogating empirical-statistical downscaling. Climatic
163, doi:10.1016/j.jhydrol.2004.10.012. Change, 122, 539–554, doi:10.1007/s10584-013-1021-z.
Eden, J. M., M. Widmann, D. Grawe, and S. Rast, 2012: Skill, Hofer, M., B. Marzeion, and T. Moelg, 2012: Comparing the skill of
correction, and downscaling of GCM-simulated precipitation. different reanalyses and their ensembles as predictors for daily
J. Climate, 25, 3970–3984, doi:10.1175/JCLI-D-11-00254.1. air temperature on a glaciated mountain (Peru). Climate Dyn.,
Fealy, R., and J. Sweeney, 2007: Statistical downscaling of pre- 39, 1969–1980, doi:10.1007/s00382-012-1501-2.
cipitation for a selection of sites in Ireland employing Kang, H., K. H. An, C. K. Park, A. L. S. Solis, and K. Stitthichivapak,
a generalised linear modelling approach. Int. J. Climatol., 27, 2007: Multimodel output statistical downscaling prediction of
2083–2094, doi:10.1002/joc.1506. precipitation in the Philippines and Thailand. Geophys. Res.
Flores, J. F., and V. F. Balagot, 1969: Climate of the Philippines. Lett., 34, L15710, doi:10.1029/2007GL030730.
Climates of Northern and Eastern Asia, Vol. 8, World Survey Kintanar, R. L., 1984: Climate of the Philippines. Philippine At-
of Climatology, H. E. Landsburg, Ed., Elsevier, 159–213. mospheric, Geophysical and Astronomical Services Admin-
Fowler, H. J., S. Blenkinsop, and C. Tebaldi, 2007: Linking climate istration (PAGASA) Tech. Rep., 38 pp.
change modelling to impacts studies: Recent advances Koukidis, E. N., and A. A. Berg, 2009: Sensitivity of the Statistical
in downscaling techniques for hydrological modelling. Int. DownScaling Model (SDSM) to reanalysis products. Atmos.–
J. Climatol., 27, 1547–1578, doi:10.1002/joc.1556. Ocean, 47, 1–18, doi:10.3137/AO924.2009.
Giorgetta, M. A., G. P. Brasseur, E. Roeckner, and J. Marotzke, Manzanas, R., L. K. Amekudzi, K. Preko, S. Herrera, and J. M.
2006: Preface to special section on climate models at the Max Gutiérrez, 2014: Precipitation variability and trends in Ghana:
Planck Institute for Meteorology. J. Climate, 19, 3769–3770, An intercomparison of observational and reanalysis products.
doi:10.1175/JCLI9023.1. Climatic Change, 124, 805–819, doi:10.1007/s10584-014-1100-9.
Giorgi, F., and L. Mearns, 2002: Calculation of average, uncertainty Maraun, D., and Coauthors, 2010: Precipitation downscaling under
range, and reliability of regional climate changes from climate change: Recent developments to bridge the gap be-
AOGCM simulations via the ‘‘reliability ensemble averaging’’ tween dynamical models and the end user. Rev. Geophys., 48,
(REA) method. J. Climate, 15, 1141–1158, doi:10.1175/ RG3003, doi:10.1029/2009RG000314.
1520-0442(2002)015,1141:COAURA.2.0.CO;2. Marzban, C., S. Sandgathe, and E. Kalnay, 2006: MOS, perfect
Goodess, C., and J. Palutikof, 1998: Development of daily prog, and reanalysis. Mon. Wea. Rev., 134, 657–663, doi:10.1175/
rainfall scenarios for southeast Spain using a circulation-type ap- MWR3088.1.
proach to downscaling. Int. J. Climatol., 18, 1051–1083, doi:10.1002/ Nelder, J. A., and R. W. M. Wedderburn, 1972: Generalized linear
(SICI)1097-0088(199808)18:10,1051::AID-JOC304.3.0.CO;2-1. models. J. Roy. Stat. Soc., 135A, 370–384, doi:10.2307/2344614.
Grotch, S. L., and M. C. MacCracken, 1991: The use of general Onogi, K., and Coauthors, 2007: The JRA-25 Reanalysis. J. Meteor.
circulation models to predict regional climatic change. Soc. Japan, 85, 369–432, doi:10.2151/jmsj.85.369.
J. Climate, 4, 286–303, doi:10.1175/1520-0442(1991)004,0286: Park, J.-H., S.-G. Oh, and M.-S. Suh, 2013: Impacts of boundary
TUOGCM.2.0.CO;2. conditions on the precipitation simulation of RegCM4 in the
Gutiérrez, J. M., D. San-Martín, S. Brands, R. Manzanas, and CORDEX East Asia domain. J. Geophys. Res. Atmos., 118,
S. Herrera, 2013: Reassessing statistical downscaling techniques 1652–1667, doi:10.1002/jgrd.50159.

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC

4184 JOURNAL OF CLIMATE VOLUME 28

Paul, S., C. M. Liu, J. M. Chen, and S. H. Lin, 2008: Development of different downscaling techniques. J. Climate, 14, 4422–4446,
a statistical downscaling model for projecting monthly rainfall doi:10.1175/1520-0442(2001)014,4422:PSOIAC.2.0.CO;2.
over East Asia from a general circulation model output. Vrac, M., M. L. Stein, K. Hayhoe, and X.-Z. Liang, 2007: A general
J. Geophys. Res., 113, D15117, doi:10.1029/2007JD009472. method for validating statistical downscaling methods under
Räisänen, J., 2007: How reliable are climate models? Tellus, 59A, future climate change. Geophys. Res. Lett., 34, L18701,
2–29, doi:10.1111/j.1600-0870.2006.00211.x. doi:10.1029/2007GL030295.
Sauter, T., and V. Venema, 2011: Natural three-dimensional pre- Widmann, M., C. S. Bretherton, and E. P. Salathé, 2003: Statistical
dictor domains for statistical precipitation downscaling. precipitation downscaling over the northwestern United
J. Climate, 24, 6132–6145, doi:10.1175/2011JCLI4155.1. States using numerically simulated precipitation as a predictor.
Sterl, A., 2004: On the (in)homogeneity of reanalysis products. J. Climate, 16, 799–816, doi:10.1175/1520-0442(2003)016,0799:
J. Climate, 17, 3866–3873, doi:10.1175/1520-0442(2004)017,3866: SPDOTN.2.0.CO;2.
OTIORP.2.0.CO;2. Wilby, R. L., and T. M. L. Wigley, 1997: Downscaling general
Timbal, B., and D. A. Jones, 2008: Future projections of winter circulation model output: A review of methods and limita-
rainfall in southeast Australia using a statistical down- tions. Prog. Phys. Geogr., 21, 530–548, doi:10.1177/
scaling technique. Climatic Change, 86, 165–187, doi:10.1007/ 030913339702100403.
s10584-007-9279-7. ——, H. Hassan, and K. Hanaki, 1998: Statistical downscaling of hy-
——, A. Dufour, and B. McAvaney, 2003: An estimate of future drometeorological variables using general circulation model
climate change for western France using a statistical down- output. J. Hydrol., 205, 1–19, doi:10.1016/S0022-1694(97)00130-3.
scaling technique. Climate Dyn., 20, 807–823, doi:10.1007/ ——, S. Charles, E. Zorita, B. Timbal, P. Whetton, and L. Mearns,
s00382-002-0298-9. 2004: Guidelines for use of climate scenarios developed from
Trenberth, K. E., D. P. Stepaniak, J. W. Hurrell, and M. Fiorino, 2001: statistical downscaling methods. Tech. Rep., IPCC TGCIA, 27 pp.
Quality of reanalyses in the tropics. J. Climate, 14, 1499–1510, ——, J. Troni, Y. Biot, L. Tedd, B. C. Hewitson, D. M. Smith, and
doi:10.1175/1520-0442(2001)014,1499:QORITT.2.0.CO;2. R. T. Sutton, 2009: A review of climate risk information for
Trigo, R., and J. Palutikof, 2001: Precipitation scenarios over adaptation and development planning. Int. J. Climatol., 29,
Iberia: A comparison between direct GCM output and 1193–1215, doi:10.1002/joc.1839.

Unauthenticated | Downloaded 07/22/24 04:58 PM UTC