Content uploaded by H. Douville
Author content
All content in this area was uploaded by H. Douville on Mar 24, 2015
Content may be subject to copyright.
Relative contribution of soil moisture and snow mass
to seasonal climate predictability: a pilot study
Herve
´Douville
Received: 26 September 2008 / Accepted: 9 December 2008 / Published online: 16 January 2009
Springer-Verlag 2009
Abstract Land surface hydrology (LSH) is a potential
source of long-range atmospheric predictability that has
received less attention than sea surface temperature (SST).
In this study, we carry out ensemble atmospheric simula-
tions driven by observed or climatological SST in which
the LSH is either interactive or nudged towards a global
monthly re-analysis. The main objective is to evaluate the
impact of soil moisture or snow mass anomalies on sea-
sonal climate variability and predictability over the 1986–
1995 period. We first analyse the annual cycle of zonal
mean potential (perfect model approach) and effective
(simulated vs. observed climate) predictability in order to
identify the seasons and latitudes where land surface ini-
tialization is potentially relevant. Results highlight the
influence of soil moisture boundary conditions in the
summer mid-latitudes and the role of snow boundary
conditions in the northern high latitudes. Then, we focus on
the Eurasian continent and we contrast seasons with
opposite land surface anomalies. In addition to the nudged
experiments, we conduct ensembles of seasonal hindcasts
in which the relaxation is switched off at the end of spring
or winter in order to evaluate the impact of soil moisture or
snow mass initialization. LSH appears as an effective
source of surface air temperature and precipitation pre-
dictability over Eurasia (as well as North America), at least
as important as SST in spring and summer. Cloud feed-
backs and large-scale dynamics contribute to amplify the
regional temperature response, which is however, mainly
found at the lowest model levels and only represents a
small fraction of the observed variability in the upper
troposphere.
1 Introduction
Over recent decades, the recognition that slowly-evolving
boundary conditions can be a source of atmospheric pre-
dictability at the seasonal timescale has promoted the
development of coupled ocean-atmosphere general circu-
lation models (GCMs) as well as the design of ocean data
assimilation techniques (Palmer and Anderson 1994). Pilot
studies such as the Development of a European Multimodel
Ensemble system for seasonal to inTERannual prediction
(DEMETER, Palmer 2005) project have demonstrated the
potential of dynamical seasonal prediction (DSP) systems,
which are now operated routinely by several countries.
Nevertheless, it has been recently suggested that our ability
to predict regional climate anomalies at the seasonal
timescale has reached a plateau (WCRP 2008). While
decadal prediction is now emerging as a new challenge for
the global climate modeling community, the improvement
of current DSP systems should not be relegated to a sec-
ondary objective.
In June 2007, the first WCRP Workshop on Seasonal
Prediction was held in Barcelona to define a road-map for
the next decade. On the one hand, it was recognized that
the relative success of the multi-model ensemble fore-
casting (Doblas-Reyes et al. 2005) should not obviate the
need of carrying on the development of coupled ocean–
atmosphere GCMs and of ocean data assimilation tech-
niques. On the other hand, it was highlighted that other
components of the global climate system could contribute
to improved forecast skill, and that ‘‘land–atmosphere
H. Douville (&)
Me
´te
´o-France/CNRM/GMGEC/UDC, 42 Avenue Coriolis,
31057 Toulouse Cedex 01, France
e-mail: herve.douville@meteo.fr
123
Clim Dyn (2010) 34:797–818
DOI 10.1007/s00382-008-0508-1
interactions are perhaps the most obvious example of the
need to improve the representation of climate system
interactions and their potential to improve forecast qua-
lity’’. While such a perspective is not new (Delworth and
Manabe 1989; Dirmeyer 2000; Douville and Chauvin
2000; Koster et al. 2000), the lack of global land surface
observations and of reliable land surface data assimilation
techniques has been and still is a major obstacle for
drawing robust conclusions on this issue.
In the mid-1990s, the International Satellite Land Sur-
face Climatology Project (ISLSCP) has been launched and
has allowed land surface modelers to produce global soil
moisture climatologies by driving their models with com-
mon atmospheric forcings in the framework of the Global
Soil Wetness Project (GSWP, http://grads.iges.org/gswp).
The Interaction Soil–Biosphere–Atmosphere (ISBA,
Mahfouf et al. 1995; Douville et al. 1995) land surface
model of Centre National de Recherches Me
´te
´orologiques
(CNRM) has contributed to GSWP and has been driven by
the ISLSCP data first from 1987 to 1988 (GSWP-1,
Douville 1998), then from 1986 to 1995 (GSWP-2,
Decharme and Douville 2007). Besides control runs using
the common ISLSCP soil and vegetation parameters, par-
allel integrations have been achieved with the native ISBA
land surface parameters to produce soil moisture and snow
mass climatologies that are fully consistent with the
CNRM atmospheric GCM. These climatologies have been
used to nudge global atmospheric simulations and compare
the relative influence of monthly soil moisture and SST on
atmospheric variability and predictability at the seasonal
timescale (Douville and Chauvin 2000; Douville 2002;
Conil et al. 2007).
Consistent with former or parallel studies (i.e., Dirmeyer
2005), these sensitivity experiments have emphasized the
relevance of soil moisture boundary conditions for cap-
turing the interannual climate variability observed at the
regional scale, particularly during the boreal summer sea-
son. Obviously, the strength and spatial distribution of the
land-surface coupling is, however, model-dependent given
the diversity of atmospheric GCMs. This issue was tackled
by the global land–atmosphere coupling experiment
(GLACE) intercomparison project (Koster and the GLACE
team 2004) aimed at comparing where and to what extent
boreal summer precipitation is controlled by soil moisture
in a dozen of models. The results showed a large spread
between the models, but highlighted three ‘‘hotspots’’
where the coupling appears as relatively strong in a
majority of models: North America, Sahel and northern
India. Nevertheless, the conclusions of GLACE should be
considered with caution for at least three reasons. First, no
observational counterpart of the coupling strength is
available to confirm this distribution. Second, the metric
that was used to measure the coupling strength was focused
on subseasonal rather than seasonal variability. Third, the
experiment design was based on seasonal hindcasts driven
by the 1994 monthly SST and the results might have been
somewhat different with another SST forcing.
The CNRM atmospheric GCM did not participate in
GLACE, but also shows a significant precipitation sensi-
tivity to soil moisture over the Sahel (Douville et al. 2001;
Douville 2002) and North America (Douville 2004; Conil
et al. 2007). In contrast with the results of GLACE, India
does not appear as a region of strong coupling due to a
negative dynamical feedback (less moisture convergence)
that cancels the positive evaporation feedback over this
region when the whole summer monsoon season is con-
sidered (Douville et al. 2001). Conversely, the CNRM
model suggests that Europe is another region of strong
coupling in summer (Douville and Chauvin 2000; Conil
et al. 2007). This result is consistent with observational and
numerical studies highlighting the potential contribution of
soil moisture deficit to heat and drought waves over
Western Europe (Ferranti and Viterbo 2006; Vautard et al.
2007), as well as with a recent statistical analysis of
soil moisture feedbacks in the CMIP3 coupled ocean-
atmosphere simulations (Notaro 2008). As far as the Sahel
is concerned, Douville et al. (2007) highlighted the fact that
the relatively strong coupling found in most GCMs
(including in the CNRM model) does not guarantee a
strong influence of soil moisture on the all-summer
monsoon precipitation due to the dominant contribution of
moisture convergence to the variability of rainfall (at least
in the first part of the rainy season) and its strong sensitivity
to the tropical SST.
Most studies do not tell much about the predictability of
soil moisture itself and our ability to improve DSP systems
through a better initialization of the LSH. To go one step
further, Conil et al. (2008) have conducted additional
ensembles of global atmospheric simulations driven by
observed SST in which the soil moisture nudging is
removed at the end of May in order to explore the impact
of soil moisture initial conditions on summer hindcasts.
Such experiments are similar to those formerly performed
by Koster et al. (2004), but the focus is not limited to North
America and to 1-month hindcasts. The results indicate that
the CNRM atmospheric GCM is better than simple (auto-
regressive) statistical models for predicting the persistence
of soil moisture anomalies. They also suggest that soil
moisture memory is able to sustain a significant atmo-
spheric predictability at the monthly to seasonal timescale.
The present article is the follow-on of Conil et al. (2007,
2008). Besides soil moisture, it also explores the influence
of the Northern Hemisphere snow cover on atmospheric
variability and predictability. Here again, former studies
have been hampered by the lack of global observed cli-
matologies (Kumar and Yang 2003). Visible imagery does
798 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
provide reliable estimates of the Northern Hemisphere
snow cover since the late 1960s, but does not give access to
snow depth anomalies. Off-line snow depth analyses have
been recently produced (Brown et al. 2003), but not on the
global scale. On-line NWP data assimilation techniques
remain relatively crude and still show serious deficiencies
when compared with in situ observations. Therefore, the
nudging towards the monthly GSWP climatology deve-
loped by Douville and Chauvin (2000) again appears as an
attractive strategy to prescribe ‘‘realistic’’ snow boundary
conditions in the CNRM atmospheric model. Moreover, as
in Conil et al. (2008), the nudging can be removed at the
beginning of a particular season to explore the relevance of
initial versus boundary conditions.
The experiment design is further described in Sect. 2.In
Sect. 3, the annual cycle of predictability is compared
between the control and nudged experiments. In Sect. 4,a
regional and seasonal analysis, including the role of initial
conditions, is conducted for particular years with contrasted
land surface anomalies. The focus is on the Eurasian conti-
nent, which does not necessarily appear as a region of strong
land-atmosphere coupling in former numerical studies. A
summary and discussion of the results is given in Sect. 5.
2 Experiment design and predictability metrics
The Arpege-Climat atmospheric GCM coupled to the ISBA
land surface model (Mahfouf et al. 1995; Douville et al.
1995) has been used to perform global seasonal hindcasts
in which soil moisture or snow mass is either interactive or
strongly relaxed towards a monthly reanalysis. The
reanalysis covers the 1986–1995 period and has been
produced by driving the ISBA model with a combination of
3-hourly atmospheric analyses and of monthly climatolo-
gies based on in situ and satellite observations (Dirmeyer
et al. 2006; Decharme and Douville 2007). In both off-line
and on-line configurations, ISBA here uses a single-layer
snow model (Douville et al. 1995) and a simple force-
restore soil hydrology with a Variable Infiltration Capacity
runoff scheme. The GSWP-2 reanalysis is first used to
nudge either soil moisture or snow mass in ensembles of
AMIP-type simulations, i.e., global atmospheric simulations
driven by observed monthly mean SST. Then climatological
rather than observed monthly mean SST are prescribed to
isolate the land surface contribution to climate variability.
Finally, the impact of land surface initialization is explored
by removing the nudging at the beginning of seasonal
hindcasts driven by observed SST.
Table 1summarizes the various experiments. Most
ensembles consist of ten integrations from 1st September
1985 to 31st December 1995. The different initial condi-
tions are derived from 10 consecutive years of a pre-
existing AMIP-type experiment. The first 4 months of each
member are considered as a spin-up and no attention is paid
to the impact of atmospheric initialization on seasonal
predictability. In the control case (FF), there is no nudging
so that soil moisture and snow mass are fully interactive
with the atmosphere. In a first pair of sensitivity experi-
ments (GG and HH), soil moisture and snow mass,
respectively, are nudged towards the GSWP-2 reanalysis
interpolated onto the model grid and averaged on a
monthly basis. Then, parallel experiments (GC and HC)
have been performed with climatological SST (mean
annual cycle averaged over the 1986–1995 period) to
evaluate the robustness of the results and the additivity of
the oceanic and land surface contributions to seasonal
predictability. Finally, two ensembles of seasonal hindcasts
using an interactive LSH and prescribed observed SST (GF
and HF) have been performed to assess the impact of initial
rather than boundary conditions of soil moisture and snow
mass. Each of the ten integrations for each of the 10 years
(1986–1995) starts from GG and HH initial conditions,
respectively, either at the end of May (GF, impact of soil
moisture initialization on boreal summer hindcasts) or at
the end of March (HF, impact of snow mass initialization
on boreal spring hindcasts).
Further technical details on the nudging technique can
be found in Douville (2003). The nudged soil moisture
experiments have been already compared to the control
ensemble by Conil et al. (2007). The influence of soil
moisture initialization on summer hindcasts has been also
Table 1 Summary of the experiments
Expt SST Soil moisture Snow mass
FF AMIP Free Free
GG AMIP Nudged toward GSWP-2 Free
HH AMIP Free Nudged toward GSWP-2
GC AMIP monthly climatology Nudged toward GSWP-2 Free
HC AMIP monthly climatology Free Nudged toward GSWP-2
GF AMIP (June to September) Initialized from GG in late May then free Free
HF AMIP (April to September) Free Initialized from HH in late March then free
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 799
123
explored by Conil et al. (2008). These results are further
discussed in the continuation of the present study and are
compared to the impacts of snow boundary conditions and
of snow initialization. Note again that none of these
experiments deals with the impact of atmospheric initiali-
zation, which however, can be important during the first
month of the simulations. Such an experiment design
allows us to focus only on the impact of the land surface
initialization in the GF and HF experiments.
Two metrics have been used to quantify predictability in
the various experiments. We have first conducted a one-
way analysis of variance (ANOVA) to evaluate potential
predictability (hereafter PP) and its relative sensitivity to
LSH and SST boundary conditions. This technique allows
us to split the total variance of the CNRM model into a
chaotic internal component and an external component
driven by the lower boundary conditions (Douville 2004).
Details on the methodology and its underlying hypotheses
can be found in Von Storch and Zwiers (1999). In this
perfect-model approach, PP is the ratio of external versus
total variability. The total variance is estimated from the
100 seasonal integrations each experiment consists of,
while the externally forced variance ignores the contribu-
tion of the atmospheric initial conditions and is computed
from the ten ensemble mean seasons (from 1986 to 1995).
This idealized framework is here justified by the use of
soil moisture and snow mass boundary conditions that are not
derived directly from observations but are calculated by a
land surface model. The gap between potential and effective
predictability is therefore not entirely due to model defi-
ciencies, but also to the inherent limitations of the GSWP-2
climatology. Effective predictability or model skill has been
here simply measured as the temporal Anomaly Correlation
Coefficient (ACC) between the simulated climate anomalies
and those from the Climate Research Unit version 2 (CRU2,
http://www.cru.uea.ac.uk/cru/data/) climatology after inter-
polation onto the GCM horizontal grid.
Given the 10-year framework of the study, sucha skill score
is obviously not robust. Comparing the PP and ACC distri-
butions is thus important to distinguish between spurious
and genuine peaks of effective predictability. A common
assumption is indeed that PP represents an upper limit of
effective predictability. While this hypothesis is not neces-
sarily valid (since there is no guarantee that the signal to noise
ratio is correct in the model), high PP is a necessary condition
for the model to show someskill and PP is therefore very useful
to detect high ACCs which are due to a stochastic artefact.
3 Annual cycle of zonal mean predictability
Another way to reduce the stochastic noise associated with
the limited 10-year sampling is to show zonal mean rather
than grid-point values of predictability. This is done in
Fig. 1which shows the impact of the nudging on the
annual cycle of total variability, PP and ACC for surface
air temperature over land areas comprised between 55S
and 75N. Not surprisingly, the control experiment shows a
maximum variability in the winter extratropics and mini-
mum values in the Tropics. The distribution is not much
sensitive to the nudging and only shows a weak decrease in
total variability when using climatological rather than
observed SST.
Moving to PP, the control experiment shows as expected
maximum predictability in the Tropics where SST exerts a
strong impact on atmospheric variability. Large values are
also found in the Southern Hemisphere due to the limited
number of land grid points and to the strong SST forcing in
coastal regions. Conversely, PP remains all year round
fairly limited in the northern extratropics. This feature is
however very sensitive to the land surface boundary con-
ditions. The soil moisture nudging (GG) increases PP in the
boreal summer mid-latitudes, while the snow mass nudging
(HH) has a similar impact during the springtime snowmelt
and, though to lesser extent, at the beginning of the snow
season. Such a sensitivity is confirmed by the results of GC
and HC, which show consistent PP signals in the northern
extratropics despite the use of climatological SST. This
result suggests that the SST and LSH forcings are relatively
additive in our experiments for these regions. Note how-
ever that both sources of atmospheric variability are not
necessarily independent in the real world.
The last column in Fig. 1shows the annual cycle of zonal
mean effective predictability. In line with the PP distribu-
tion and with the results of operational DSP systems,
the control experiment shows maximum skill scores in the
Tropics. The ACC distribution is relatively noisy in the
extratropics, but shows correlations that are generally less
than 0.3. The nudged experiments show patches of higher
predictability in which we are relatively confident given
their consistency with the response of PP. The relevance of
soil moisture boundary conditions (GG and GC) in the
boreal summer mid-latitudes is confirmed. The impact of
snow boundary conditions (HH and HC) in the northern
extratropics is also found, but is apparently stronger in fall
and winter than in spring. This apparent discrepancy with
the response of PP might be explained by at least two fac-
tors. On the one hand, zonal mean ACC provide a biased
estimate of predictability because of the limited 10-year
period of the study. On the other hand, the spring peak of
zonal mean PP could be related to a delayed snowmelt in the
ISBA model (Decharme and Douville 2007) which could
artificially inflate the potential predictability of surface air
temperature in this particular season.
As far as land precipitation is concerned (Fig. 2), all
experiments show a robust annual cycle of interannual
800 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
0.2
0.4
0.6
0.8
1
1.5
2
3
4
Tot Var TCLS FF
JFMAMJJASOND
-30
0
30
60
0
5
10
15
20
25
30
40
50
100
Pot Pred TCLS FF
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC TCLS FF/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0.2
0.4
0.6
0.8
1
1.5
2
3
4
Tot Var TCLS GG
JFMAMJJASOND
-30
0
30
60
0
5
10
15
20
25
30
40
50
100
Pot Pred TCLS GG
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC TCLS GG/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0. 2
0.4
0. 6
0.8
1
1.5
2
3
4
Tot Var TCLS GC
JFMAMJJASOND
-30
0
30
60
0
5
10
15
20
25
30
40
50
100
Pot Pred TCLS GC
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0. 2
0.25
0. 3
0.4
0. 5
1
ACC TCLS GC/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0.2
0.4
0.6
0.8
1
1.5
2
3
4
Tot Var TCLS HH
JFMAMJJASOND
-30
0
30
60
0
5
10
15
20
25
30
40
50
100
Pot Pred TCLS HH
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC TCLS HH/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0.2
0.4
0.6
0.8
1
1.5
2
3
4
Tot Var TCLS HC
JFMAMJJASOND
-30
0
30
60
0
5
10
15
20
25
30
40
50
100
Pot Pred TCLS HC
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC TCLS HC/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
Fig. 1 Zonal mean annual cycle of land surface air temperature:
interannual standard deviation (left column, in K), potential predict-
ability (central column, in %) and effective predictability against the
CRU2 climatology (right column, dimensionless ACC) in FF, GG,
HH, GC and HC, respectively
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 801
123
0
0.1
0.2
0.3
0.5
1
1.5
2
3
Tot Var PTOT FF
JFMAMJJASOND
-30
0
30
60
0
4
8
12
16
20
25
30
40
50
Pot Pred PTOT FF
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC PTOT FF/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0
0.1
0.2
0.3
0.5
1
1.5
2
3
Tot Var PTOT GG
JFMAMJJASOND
-30
0
30
60
0
4
8
12
16
20
25
30
40
50
Pot Pred PTOT GG
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC PTOT GG/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0
0.1
0.2
0.3
0.5
1
1.5
2
3
Tot Var PTOT GC
JFMAMJJASOND
-30
0
30
60
0
4
8
12
16
20
25
30
40
50
Pot Pred PTOT GC
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC PTOT GC/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0
0.1
0.2
0.3
0.5
1
1.5
2
3
Tot Var PTOT HH
JFMAMJJASOND
-30
0
30
60
0
4
8
12
16
20
25
30
40
50
Pot Pred PTOT HH
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC PTOT HH/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
0
0.1
0.2
0.3
0.5
1
1.5
2
3
Tot Var PTOT HC
JFMAMJJASOND
-30
0
30
60
0
4
8
12
16
20
25
30
40
50
Pot Pred PTOT HC
JFMAMJJASOND
-30
0
30
60
0
0.05
0.1
0.15
0.2
0.25
0.3
0.4
0.5
1
ACC PTOT HC/obs 1986-95
JFMAMJJASOND
EQ
40N
20S
20N
40S
60N
Fig. 2 Zonal mean annual cycle of land precipitation: interannual
standard deviation (left column, in mm/day), potential predictability
(central column, in %) and effective predictability against the CRU2
climatology (right column, dimensionless ACC) in FF, GG, HH, GC
and HC, respectively
802 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
variability with maximum values following the seasonal
migration of the inter-tropical convergence zone (ITCZ).
In the control experiment, PP is found in the Tropics,
but is much less than for surface air temperature and is
very weak in the extratropics. Nudging soil moisture
(snow mass) leads to a significant but limited increase in
PP in the boreal summer mid-latitudes (boreal spring
mid- and-high latitudes), that is found with or without
SST forcing. In line with the low values of PP, the zonal
mean ACC distribution is too noisy to get a robust
evaluation of the effective predictability. Nonetheless, a
peak of predictability appears in the boreal summer mid-
latitudes when the soil moisture nudging is activated.
This maximum is even more pronounced in GC than in
GG, suggesting that the summer SST forcing not only
does not represent a source of effective predictability
in this region, but could even exert a spurious influence
on the precipitation variability in the CNRM model
due to a poor simulation of tropical-extratropical SST
teleconnections.
4 Seasonal contrasts over Eurasia
We now concentrate on contrasted seasons, when the
potential contribution of LSH to seasonal climate predict-
ability is presumably the strongest over the 1986–1995
period. In addition to the former experiments with pre-
scribed land surface boundary conditions, the impact of
land surface initialization is also explored. The focus is on
Eurasia that contributes to the zonal belt of increased
predictability in Figs. 1and 2. Despite its larger area,
Eurasia has been less emphasized than North America as a
region of strong soil moisture feedback. Moreover, it shows
an extensive winter snow cover which is also likely to
contribute to seasonal predictability.
1986 1988 1990 1992 1994 1996
YEARS
-30
-20
-10
0
10
20
30
WP anomaly (kg/m2)
JJA WP anomalies - FF - R=-0.12
CENTRAL EUROPE [45N-65N,8E-48E]
1986 1988 1990 1992 1994 1996
YEARS
-30
-20
-10
0
10
20
30
WP anomaly (kg/m2)
JA WP anomalies - FF - R=-0.19
CENTRAL EUROPE [45N-65N,8E-48E]
1986 1988 1990 1992 1994 1996
YEARS
-30
-20
-10
0
10
20
30
WP anomaly (kg/m2)
JJA WP anomalies - GF - R=0.68
CENTRAL EUROPE [45N-65N,8E-48E]
1986 1988 1990 1992 1994 1996
YEARS
-30
-20
-10
0
10
20
30
WP anomaly (kg/m2)
JA WP anomalies - GF - R=0.40
CENTRAL EUROPE [45N-65N,8E-48E]
Fig. 3 June to August (JJA, left panels) and July to August (JA, right
panels) soil moisture anomalies (kg/m
2
) averaged over Central
Europe from 1986 to 1995. Ensemble mean anomalies (black
squares) simulated in the control (FF, upper panels) and seasonal
hindcast (GF, lower panels) experiments are compared to GSWP-2
anomalies (red disks). For each experiment, R is the 10-year
correlation of the ensemble mean values with the observations,
triangles show ±1 standard deviation and solid lines correspond to
the minimum and maximum anomalies
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 803
123
4.1 Predictability of soil moisture and snow mass
A first condition to be filled for the lower boundary
conditions to play a role in seasonal climate prediction is
the predictability of this lower boundary forcing itself. A
second related condition is the presence of anomalies with
a sufficient magnitude and spatial extent in the initial
conditions. For this reason, the predictability of the LSH
is here illustrated by focusing on two specific regions:
Central Europe [45–65N/8–48E] for summer soil
moisture and Central Eurasia [40–70N/50–90E] for
spring snow mass.
Figures 3and 4compare the ensemble mean anomalies
simulated over these regions in the control (FF) and sea-
sonal hindcast (GF or HF) experiments to the GSWP-2
anomalies, relative to their respective 1986–1995 clima-
tology. While again a 10-year period is too short to derive
robust correlations, the contrasted spread and predictability
found between FF on the one hand and GF or HF on the
other hand is sufficient to draw the following conclusions.
Firstly, the predictability of the LSH is very poor in the
control experiment. This result indicates that the SST
forcing is weak or even wrong (ex: summer 1992) in the
model. Secondly, the spread is clearly reduced and the
correlation is strongly improved in the seasonal hindcasts,
suggesting a significant predictability of soil moisture and
snow mass at the seasonal timescale. This predictability is
also found when focusing on months 2–3 (right panels) and
is therefore not limited to the first month after initialization.
Such a persistence could be even more obvious in a more
realistic set-up including analysed rather than random
atmospheric initial conditions.
Figures 3and 4allow us to select two pairs of summer
(1987 vs. 1992) and spring (1993 vs. 1995) seasons to
1986 1988 1990 1992 1994 1996
YEARS
-20
-10
0
10
20
WN anomaly (kg/m2)
AMJ WN anomalies - FF - R=-0.32
CENTRAL EURASIA [40N-70N,50E-90E]
1986 1988 1990 1992 1994 1996
YEARS
-20
-10
0
10
20
WN anomaly (kg/m2)
MJ WN anomalies - FF - R=-0.36
CENTRAL EURASIA [40N-70N,50E-90E]
1986 1988 1990 1992 1994 1996
YEARS
-20
-10
0
10
20
WN anomaly (kg/m2)
AMJ WN anomalies - HF - R=0.88
CENTRAL EURASIA [40N-70N,50E-90E]
1986 1988 1990 1992 1994 1996
YEARS
-20
-10
0
10
20
WN anomaly (kg/m2)
MJ WN anomalies - HF - R=0.82
CENTRAL EURASIA [40N-70N,50E-90E]
Fig. 4 April to June (AMJ, left panels) and May to June (MJ, right
panels) snow mass anomalies (kg/m
2
) averaged over Central Eurasia
from 1986 to 1995. Ensemble mean anomalies (black squares)
simulated in the control (FF, upper panels) and seasonal hindcast (HF,
lower panels) experiments are compared to GSWP-2 anomalies (red
disks). For each experiment, R is the 10-year correlation of the
ensemble mean values with the observations, triangles show ±1
standard deviation and solid lines correspond to the minimum and
maximum anomalies
804 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
illustrate the impact of soil moisture and snow mass on
climate predictability. For soil moisture, 1992 minus 1987
(left panels in Fig. 5) shows a strong deficit from June to
August (JJA) over Europe that has been already empha-
sized by Conil et al. (2007,2008). While such a deficit is
not reproduced in the control experiment in spite of the
observed SST forcing, it is perfectly imposed in GG and
GC due to the nudging towards the GSWP-2 climatology.
It is also partly captured in GF showing the predictability
of soil moisture at the seasonal timescale when GSWP-2 is
only used to initialize the model at the end of May. As far
as snow is concerned, 1995 minus 1993 (right panels in
Fig. 5 Left June to August 1992 minus 1987 differences of soil
moisture (kg/m
2
) in FF, GC, GG and GF, respectively. Right April to
June 1995 minus 1993 differences of snow mass (kg/m
2
) in FF, HC,
HH and HF, respectively. For each experiment, R is the spatial
correlation with the upper panel (GSWP-2 reanalysis)
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 805
123
Fig. 5) shows a strong deficit from April to June (AMJ)
over central Eurasia, which is not found in the control
experiment, correctly imposed in the nudged experiments
(HC and HH) and relatively predictable when GSWP-2 is
just used to initialize the model.
Such contrasted seasons are therefore particularly
interesting to look at the land surface influence on
atmospheric predictability. Note however that we do not
claim that the land surface boundary conditions are really
perfect in the nudged experiments: they are just as good as
the GSWP-2 reanalysis, but they are perfectly consistent
with the CNRM atmospheric GCM given the common use
of the ISBA land surface model in the off-line GSWP-2
simulations and in the present study.
Fig. 6 June to August 1992 minus 1987 differences of surface latent (left) and sensible (right) heat fluxes (downward is positive, W/m
2
) in FF,
GC, GG and GF, respectively. For each experiment, R is the spatial correlation with the upper panel (GSWP-2 reanalysis)
806 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
4.2 JJA 1992 minus JJA 1987
Looking first at the impact of soil moisture in summer,
Fig. 6compares the differences in the latent/sensible heat
flux simulated in FF, GC, GG and GF with GSWP-2 (after
interpolation onto the GCM horizontal grid). Not surpris-
ingly, the soil moisture deficit is associated with a strong
change in the Bowen ratio over Europe that is well captured
in the nudged experiments (GC and GG) and also relatively
consistent with GSWP-2 in the seasonal hincasts (GF).
Figure 7compares the simulated differences in net surface
shortwave and longwave radiation against the International
Satellite Cloud Climatology Project (ISCCP). This dataset
indicates that the 1992 minus 1987 soil moisture deficit is
Fig. 7 June to August 1992 minus 1987 differences of surface net shortwave (left) and net longwave (right) radiation (downward is positive, W/
m
2
) in FF, GC, GG and GF, respectively. For each experiment, R is the spatial correlation with the upper panel (ISCCP2 climatology)
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 807
123
associated with an increase in net shortwave radiation and a
decrease in net longwave radiation. Such anomalies are
captured in the nudged experiments and are still visible in
the seasonal hindcasts. They are associated with a signifi-
cant decrease in total cloud cover which is indeed observed
(not shown) in the ISCCP climatology and in the ERA40
reanalysis (Uppala et al. 2005).
In line with the response of the surface energy budget,
the CRU2 climatology shows a strong warming and a
precipitation deficit over Europe (Fig. 8). The temperature
anomalies are weak and poorly simulated in the control
experiment. In contrast, the warming is correctly simulated
in the nudged experiments. The seasonal hindcasts also
show a significant improvement compared to the control
Fig. 8 June to August 1992 minus 1987 differences of surface air temperature (left, K) and precipitation (right, mm/day) in FF, GC, GG and GF,
respectively. For each experiment, R is the spatial correlation with the upper panel (CRU2 climatology)
808 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
case, thereby emphasizing the benefit from a better ini-
tialization of soil moisture. The same conclusions apply to
the precipitation differences, even if the spatial correlation
with the observed anomalies is systematically lower than
for surface air temperature given the stronger magnitude
and finer scale of precipitation variability.
As emphasized in Douville and Chauvin (2000) and
Conil et al. (2007), the temperature sensitivity to the soil
moisture forcing is not confined to the land surface and the
precipitation sensitivity is not limited to a simple evapo-
ration feedback. Figure 9shows the simulated differences
in temperature and wind vectors at 850 hPa, as well as the
Fig. 9 June to August 1992 minus 1987 differences of temperature at
850 hPa (left, K) and geopotential eddy component at 500 hPa (right,
m) in FF, GC, GG and GF, respectively. For each experiment, R is the
spatial correlation with the upper panel (ERA40 reanalysis). Left
panels horizontal wind vector anomalies at 850 hPa are superimposed
on temperature differences. Right panels black contours denote the
±5m isolines and shading show simulated differences that are
statistically significant at a 5% level
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 809
123
eddy component of the 500 hPa geopotential height
(Z500* which is obtained after removing the zonal mean
values). The comparison of GC and FF first suggests that
the soil moisture forcing is as strong as the SST forcing in
the free troposphere (above the planetary boundary layer).
The comparison with the ERA40 reanalysis also confirms
our former hypothesis whereby the SST forcing could have
a spurious influence on the extratropical circulation simu-
lated by the CNRM atmospheric GCM. The maximum
correlation with ERA40 is obtained in GC rather than in
GG, emphasizing the dominant and more useful soil
moisture forcing in the model. Moreover, the results
Fig. 10 July–August (JA, left) and August–September (AS, right) 1992 minus 1987 differences of surface air temperature (K) in FF, GC, GG
and GF, respectively. For each experiment, R is the spatial correlation with the upper panel (CRU2 climatology)
810 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
suggest a large-scale dynamical response to the regional
soil moisture deficit that is also found in the seasonal
hindcasts but is underestimated in all experiments. This
underestimation might be due to the smoothing effect of
the ensemble mean calculation (only the predictable
counterpart of the observed pattern is shown), but could
also denote a lack of sensitivity of the CNRM model.
Figure 10 comes back to the differences in surface air
temperature, but the focus is on July–August and August–
September. It first indicates that soil moisture remains a
major boundary forcing during the whole summer season
since the temperature signal remains strong in GC and GG
compared to the JJA differences shown in Fig. 8. It also
demonstrates that soil moisture initialization is relevant not
Fig. 11 April to June 1995 minus 1993 differences of surface latent (left) and sensible (right) heat fluxes (downward is positive, W/m
2
) in FF,
HC, HH and HF, respectively. For each experiment, R is the spatial correlation with the upper panel (GSWP-2 reanalysis)
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 811
123
only at the monthly timescale, but also at months 2–3 or 3–
4, since the temperature signal vanishes only at the end of
the 4-month hindcast in GF, in line with the predictability
of soil moisture shown in Fig. 3.
Finally, it should be noticed that summer 1992 follows
the 1991 Pinatubo eruption. While the indirect radiative
impact of the eruption (i.e., mainly through the SST
boundary conditions) has been considered in our experi-
ments, volcanic aerosols and their direct impacts on
radiation and water cycle have been neglected. The model
is therefore not supposed to fully capture the 1992 minus
1987 contrast in temperature and precipitation.
Fig. 12 April to June 1995 minus 1993 differences of surface net shortwave (left) and net longwave (right) radiation (downward is positive, W/
m
2
) in FF, HC, HH and HF, respectively. For each experiment, R is the spatial correlation with the upper panel (ISCCP2 climatology)
812 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
4.3 AMJ 1995 minus AMJ 1993
Moving to the contrasted snow cover observed over central
Eurasia between spring 1993 and 1995, Fig. 11 first com-
pares the seasonal mean differences in the latent/sensible
heat flux simulated in FF, HC, HH and HF. The GSWP-2
reanalysis shows significant anomalies south of the maxi-
mum snow deficit, that is in the transition zone between
snow-free and snow-covered areas. In this region, an earlier
retreat of the winter snow cover is associated with a
Fig. 13 April to June 1995 minus 1993 differences of temperature
(K) at 2 m (left) and 850 hPa (right) in FF, HC, HH and HF,
respectively. For each experiment, R is the spatial correlation with the
upper panel (CRU2 climatology or ERA40 reanalysis). Left panels
horizontal wind vector anomalies at 850 hPa are superimposed on
temperature differences
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 813
123
springtime soil moisture deficit (not shown), which leads to
positive (negative) latent (sensible) heat flux anomalies.
Such anomalies are captured in the nudged experiments,
but not in the seasonal hindcasts. In the northern region, the
nudged experiments show an increase in surface
evaporation, which is much more pronounced than in the
GSWP-2 reanalysis. It compensates for a decrease in
snowmelt (not shown) which is strongly overestimated in
the nudged experiments because of the artificial imbalance
introduced in the snow mass budget.
Fig. 14 April to June 1989 minus 1988 differences of snow mass
(left, kg/m
2
) surface air temperature (middle, K) and 500 hPa
geopotential height (right, m) over North America in FF, HC, HH
and HF, respectively. For each experiment, R is the spatial correlation
with the upper panel (GSWP2, CRU2, and ERA40, respectively). For
geopotential height, black contours denote the ±5m isolines and
shading show simulated differences that are statistically significant at
a 5% level
814 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
Moving to surface radiation (Fig. 12), the ISCCP cli-
matology shows increased net solar radiation in the region
where the earlier retreat of the snow cover leads to a strong
decrease in surface albedo. Such anomalies are well
reproduced in the nudged experiments, but much weaker in
the seasonal hindcasts. In line with the associated soil
moisture deficit (not shown), the simulated increase in net
solar radiation is reinforced by a decrease in total cloud
cover that is also found in the ISCCP climatology (not
shown). The warming at the land surface and the reduced
cloudiness induce a radiative cooling that is relatively
consistent between ISCCP and the nudged experiments.
Again, the signature is much weaker in the seasonal
hindcasts, in line with the predicted snow mass deficit that
is confined to the northern latitudes (Fig. 5).
Figure 13 compares the simulated and observed differ-
ences in temperature both at 2 m and 850 hPa. Consistent
with the response of the surface energy budget, the CRU
climatology and ERA40 reanalysis show a strong warming
over northern and central Eurasia. In contrast with the
control experiment, the nudged experiments are able to
capture the surface warming, but the signal disappears at
850 hPa in the northern part of the domain where the
presence of snow in spring leads to a (too?) strong strati-
fication of the lower troposphere in the model. Moreover,
the surface warming vanishes in the seasonal hindcasts
(HF) despite the relative persistence of the initial snow
mass anomalies north of 60N (Fig. 5). Finally, the nudged
experiments with climatological SST (HC) show a wind
response at 850 hPa that is somewhat consistent with the
ERA40 reanalysis and suggest that persistent snow anom-
alies are likely to exert a significant influence on the large-
scale dynamics.
In summary, the results show a significant climate
influence of snow mass anomalies at the regional scale.
The strongest signals are found in the transition zone
between snow-free and snow-covered areas and are partly
associated with a hydrological snow effect whereby an
early snowmelt leads to a soil moisture deficit in spring. A
weak dynamical response and a significant cloud feedback
are simulated in the nudged experiments that are somewhat
consistent with the observations. In contrast with soil
moisture anomalies, no impact is noted on precipitation
(not shown) and no clear signal is found in the seasonal
hindcast experiments.
This rather pessimistic conclusion about the snow con-
tribution to seasonal predictability must be tempered. More
encouraging results have been found over North America,
for example for the difference between spring 1988 and
1989 (Fig. 14). In this case, the hindcast experiments show
surprisingly the best correlations with the observed dif-
ferences in both surface air temperature and mid-
tropospheric geopotential height. This is not only true for
the whole AMJ season, but also for each individual month
(not shown). Note however, that the predictability found in
the 1989 minus 1988 differences mainly originates from
the 1989 anomalies (relative to the 1986–1995 climato-
logy), while the snow nudging has a weaker impact in
spring 1988 (in line with the weaker snow anomalies).
Winter 1988/1989 experienced a large shift in the northern
hemisphere extratropical circulation that has been partly
attributed to snow cover anomalies observed in early
winter over Eurasia (Watanabe and Nitta 1998). Our results
(HC and HH vs. FF) are consistent with this study and
suggest a strong Eurasian snow forcing of the winter 1988/
1989 extratropical circulation (not shown). The atmo-
spheric response found in spring 1989 over North America
may therefore be related not only to the regional snow
anomalies, but also to a remote influence of the winter
Eurasian snow cover. Such teleconnexions induced by
snow anomalies have been documented both in winter (i.e.,
Cohen and Entekhabi 1999) and summer (i.e., Peings and
Douville 2008), and several mechanisms have been pro-
posed for the delay between the snow anomalies and their
atmospheric impact: a tropospheric-stratospheric pathway
involving the excitation of Rossby waves and their inter-
action with the mean flow from fall to winter, a land
hydrology pathway involving soil moisture anomalies and
their impact on the land–sea temperature contrast from
spring to summer. Such hypotheses are still a matter of
debate and are beyond the scope of the present study.
5 Summary and discussion
Seasonal climate predictability and seasonal forecasting
rely on the atmospheric sensitivity to its lower (ocean and
land) and upper (stratosphere) boundary forcings and on
the possible predictability of these boundary forcings. The
development of operational DSP systems has been moti-
vated by the increasing ability of coupled ocean-
atmosphere GCMs to simulate and predict El Nin
˜o
Southern Oscillation (ENSO) events in the tropical Pacific
and their global teleconnections. Nevertheless, the sporadic
nature of the ENSO events and their limited impact on
extratropical climate are major obstacles for providing
useful seasonal forecasts on a regular and global basis.
Moreover, many coupled GCMs still show serious defi-
ciencies in simulating realistic ENSO teleconnections (Joly
et al. 2007). Besides improving the coupled ocean–atmo-
sphere models, it is therefore important to look for other
potential sources of long-range atmospheric predictability.
Though land surface anomalies are generally less per-
sistent than tropical SST anomalies, they have significant
impacts on regional climate that are presumably easier to
capture than the remote SST effects. The GLACE
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 815
123
intercomparison project has been a first stage in the eval-
uation of the land–atmosphere coupling and its potential
contribution to climate variability. Nevertheless, the focus
was on subseasonal rather than seasonal timescale, mainly
on soil moisture and solely on the boreal summer season.
There is therefore a need for exploring the land surface
contribution to seasonal predictability in a more compre-
hensive and systematic way.
This is the reason why GLACE will be followed by an
ambitious GLACE-2 project (http://glace.gsfc.nasa.gov)
that is supported by WCRP (2008). The aim is to analyse
the impact of GSWP-2 versus random initial land surface
conditions in ensembles of 2-month atmospheric hindcasts
over the 1986–1995 summer seasons. The experiment
design is therefore close to the one used in the present
study. We recognize the need of a multi-model assessment
of land surface initialization for improving seasonal fore-
casting. Nevertheless, it is also important to conduct
parallel pilot studies in order to analyse the results of a
particular model in more detail.
The present study compares the relative contribution of
soil moisture and snow mass to seasonal climate predict-
ability. Though the focus was mainly on the Eurasian
continent, similar results were obtained over North America
and the main findings can be summarized as follows:
•Soil moisture and snow mass anomalies can be
predicted a few months ahead when they show a
sufficient magnitude and spatial extent.
•The contribution of soil moisture to boreal summer
predictability is not only found over North America, but
also over Europe in line with recent observational and
numerical studies.
•The contribution of snow cover to boreal spring
predictability is also potentially relevant, but is less
clear than for soil moisture and deserves further
analysis given the possible remote impacts of snow
anomalies.
•Both contributions should not be neglected given the
weak (though possibly underestimated) SST contribu-
tion to extratropical predictability in current atmospheric
GCMs.
•Both contributions are not confined to simple changes
in surface evaporation (soil moisture) or surface albedo
(snow), but involve changes in the various components
of the surface energy budget.
•Both contributions could also involve large-scale
dynamics and cloud feedbacks that deserve more
detailed validation studies.
While such results claim for a better land surface ini-
tialization in operational DSP systems, they need to be
confirmed by the GLACE-2 project. Moreover, several
additional issues have to be explored. First of all, soil
moisture and snow depth are not the only land surface
variables that represent a potential source of long-range
predictability. Recently, subsurface soil temperature was
found to increase surface air temperature variability and
memory, but with a negligible impact on predictability in
many regions of the world, particularly during boreal
summer season (Mahanama et al. 2008). Vegetation is also
likely to amplify climate variability at least at the multi-
decadal timescale, but its role at the seasonal timescale is
still uncertain and deserves further statistical (i.e., Liu et al.
2006) and numerical (i.e., Gao et al. 2008) studies. Finally,
floodplains or groundwaters also show a significant low-
frequency variability that could have regional impacts on
interannual climate variability (Bierkens and van den Hurk
2007), but have not yet been included in most LSMs and
have to be parametrized in a sufficiently robust way to be
coupled with global atmospheric GCMs (i.e., Decharme
et al. 2008).
The lack of observational data and the current limita-
tions of land surface data assimilation systems is another
important issue (see for instance Houser et al. 2004 for a
review about terrestrial data assimilation). For this reason,
the forcing of LSMs with meteorological analyses remains
an interesting strategy to produce land surface reanalyses.
It should be however, emphasized that GSWP-2 represents
an upper limit of what can be done routinely given the
difficulty to get accurate real-time precipitation analyses in
many regions of the world. Current efforts are devoted to
assimilate satellite data in LSMs and/or NWP models and
should provide improved global high-resolution soil
moisture and snow mass datasets in the near future. Never-
theless, it will be necessary to wait still for many years
before testing how useful such products are for under-
standing climate variability and initializing dynamical
seasonal forecasts. Meanwhile, land surface reanalyses
such as our GSWP dataset could be extended to a longer
period (using for example the Princeton atmospheric
forcings proposed by Sheffield et al. 2006) to repeat our
seasonal hindcast experiments and get a more robust
assessment of the impact of land surface initialization on
model skill.
Other issues are related to the fact that most sensitivity
studies aimed at exploring the land surface influence on
climate variability have been based on atmospheric GCMs
driven by prescribed SST. On the one hand, such an
experiment design assumes that ocean variability is gene-
rally considered to be insensitive to land surface
variability. This hypothesis is only valid if the land surface
contribution to atmospheric variability is a second-order
effect or at least confined to the continental areas, which is
still a matter of debate (i.e., Hu et al. 2004). On the other
hand, the use of prescribed SST can lead to an overesti-
mation of atmospheric sensitivity to land surface
816 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
perturbations given the fact that possible negative SST
feedbacks are thereby neglected (i.e., Douville 2005). This
is the reason why the comparison of coupled ocean-atmo-
sphere hindcasts to atmospheric only hindcasts driven by
observed SST would be valuable within the on-going
GLACE-2 intercomparison project.
Acknowledgments This work was supported by the ENSEMBLES
European project (contract GOCE-CT-2003-505539). Most 2D fig-
ures have been prepared using the GrADS software. The author wish
to thank the anonymous reviewers for their helpful comments. Thanks
are also due to Sophie Tyteca and Sebastien Conil for technical
support.
References
Bierkens MFP, van den Hurk BJJ (2007) Groundwater convergence
as a possible mechanism for multi-year persistence in rainfall.
Geophys Res Lett 34:L02402
Brown RD, Brasnett B, Robinson D (2003) Gridded North American
monthly snow depth and snow equivalent for GCM evaluation.
Atmos Ocean 41:1–14
Cohen J, Entekhabi D (1999) Eurasian snow cover variability and
Northern Hemisphere climate predictability. Geophys Res Lett
26:345–348. doi:10.1029/1998GL900321
Conil S, Douville H, Tyteca S (2007) The relative role of soil
moisture and SST in climate variability explored within ensem-
bles of AMIP-type simulations. Clim Dyn 28:125–145. doi:
10.1007/s00382-006-0172-2
Conil S, Douville H, Tyteca S (2008) Contribution of realistic soil
moisture initial conditions to boreal summer predictability. Clim
Dyn. doi:10.1007/s00382-008-0375-9
Decharme B, Douville H (2007) Global validation of the ISBA sub-
grid hydrology. Clim Dyn 29:21–37. doi:10.1007/s00382-006-
0216-7
Decharme B, Douville H, PrigentC, Papa F, Aires F (2008) A new global
river flooding scheme: off-line validation over South America.
J Geophys Res 113:D11110. doi:10.1029/2007JD009376
Delworth T, Manabe S (1989) The influence of soil wetness on near-
surface atmospheric variability. J Clim 2:1447–1462
Dirmeyer P (2000) Using a global soil wetness dataset to improve
seasonal climate simulation. J Clim 13:2900–2922. doi:10.1175/
1520-0442(2000)013\2900:UAGSWD[2.0.CO;2
Dirmeyer P (2005) The land surface contribution to the potential
predictability of boreal summer season climate. J Hydrometeorol
6:618–632. doi:10.1175/JHM444.1
Dirmeyer PA, Gao X, Zhao M, Guo Z, Oki T, Hanasaki N (2006)
GSWP-2: multi-model analysis and implications for our percep-
tion of the land surface. Bull Am Metab Soc 87:1381–1397. doi:
10.1175/BAMS-87-10-1381
Doblas-Reyes FJ, Hagedorn R, Palmer TN (2005) The rationale
behind the succes of multi-model ensembles in seasonal
forecasting. Part II: Calibration and combination. Tellus
57A:234–252
Douville H (1998) Validation and sensitivity of the global hydrologic
budget in stand-alone simulations with the ISBA land surface
scheme. Clim Dyn 14:151–171. doi:10.1007/s003820050215
Douville H (2002) Influence of soil moisture on the Asian and African
monsoons. Part II: interannual variability. J Clim 15:701–720
Douville H (2003) Assessing the influence of soil moisture on seasonal
climate variability with AGCMs. J Hydrometeorol 4:1044–1066.
doi:10.1175/1525-7541(2003)004\1044:ATIOSM[2.0.CO;2
Douville H (2004) Relevance of soil moisture for seasonal atmo-
spheric predictions: is it an initial value problem. Clim Dyn
22:429–446. doi:10.1007/s00382-003-0386-5
Douville H (2005) Limitations of time-slice experiments for predict-
ing regional climate change over South Asia. Clim Dyn 24:373–
391. doi:10.1007/s00382-004-0509-7
Douville H, Chauvin F (2000) Relevance of soil moisture for seasonal
climate predictions: a preliminary study. Clim Dyn 16:719–736.
doi:10.1007/s003820000080
Douville H, Royer J-F, Mahfouf J-F (1995) A new snow parameter-
ization for the Me
´te
´o-France climate model. Part-I: Validation in
stand-alone experiments. Clim Dyn 12:21–35. doi:10.1007/
BF00208760
Douville H, Chauvin F, Broqua H (2001) Influence of soil moisture on
the Asian and African monsoons. Part I: Mean monsoon and
daily precipitation. J Clim 14:2381–2403
Douville H, Conil S, Tyteca S, Voldoire A (2007) Soil moisture
memory and West African monsoon predictability: artefact or
reality? Clim Dyn 28:723–742. doi:10.1007/s00382-006-0207-8
Ferranti L, Viterbo P (2006) The European summer of 2003:
sensitivity to soil water initial conditions. J Clim 19:3659–
3680. doi:10.1175/JCLI3810.1
Gao X, Dirmeyer PA, Guo Z, Zhao M (2008) Sensitivity of land
surface simulations to the treatment of vegetation properties and
the implications for seasonal climate prediction. J Hydromet
9:348–366. doi:10.1175/2007JHM931.1
Houser P, Hutchinson MF, Viterbo P, Douville H, Running SW
(2004) Terrestrial data assimilation. Chapter C.4 of the BAHC
synthesis book ‘‘Vegetation, Water, Humans and the Climate’’.
Springer, Heidelberg, 545p
Hu Z-Z, Schneider EK, Bhatt US, Kirtman BP (2004) Potential
mechanism for response of El Nin
˜o-Southern Oscillation vari-
ability to change in land surface energy budget. J Geophys Res
109:D21113. doi:10.1029/2004JD004771
Joly M, Voldoire A, Douville H, Terray P, Royer J-F (2007) African
monsoon teleconnections with tropical SST in a set of IPCC4
coupled models. Clim Dyn 29:1–20. doi:10.1007/s00382-006-
0215-8
Koster R, The GLACE team (2004) Regions of strong coupling
between soil moisture and precipitation. Science 305:1138–
1140. doi:10.1126/science.1100217
Koster R, Suarez M, Heiser M (2000) Variability and predictability
of precipitation at seasonal to interannual timescales. J Hydro-
meteorol 1:26–46. doi:10.1175/1525-7541(2000)001\0026:
VAPOPA[2.0.CO;2
Koster R, Suarez M, Liu P, Jambor U, Berg A, Kistler M, Reichle R,
Rodell R, Famiglietti J (2004) Realistic initialization of land
surface states: impacts on subseasonal forecast skill. J Hydro-
meteorol 5:1049–1063. doi:10.1175/JHM-387.1
Kumar A, Yang F (2003) Comparative influence of snow and SST
variability on extratropical climate in northern winter. J Clim
16:2248–2261. doi:10.1175/2771.1
Liu Z, Notaro M, Kutzbach J (2006) Assessing global vegetation-
climate feedbacks from observations. J Clim 19:787–814
Mahanama SPP, Koster RD, Reichle RH, Suarez MJ (2008) Impact of
subsurface temperature variability on surface air temperature
variability: an AGCM study. J Hydromet 9:804–815. doi:
10.1175/2008JHM949.1
Mahfouf J-F, Manzi AO, Noilhan J, Giordani H, De
´que
´M (1995) The
land surface scheme ISBA within the Me
´te
´o-France climate
model ARPEGE. Part I: Implementation and preliminary results.
J Clim 8:2039–2057. doi:10.1175/1520-0442(1995)008\2039:
TLSSIW[2.0.CO;2
Notaro M (2008) Statistical identification of global hot spots in soil
moisture feedbacks among IPCC AR4 models. J Geophys Res
113:D09101. doi:10.1029/2007JD009199
H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability 817
123
Palmer T (2005) Preface of the special issue of Tellus on the
DEMETER European project. Tellus 57A:217–218
Palmer T, Anderson DLT (1994) The prospect for seasonal forecast-
ing—a review paper. Quaterly J R Metab Soc 120:755–793
Peings Y, Douville H (2008) Influence of the Northern Hemisphere
snow cover on the Indian summer monsoon variability in
observations and CMIP3 simulations. Clim Dyn (submitted)
Sheffield J, Goteti G, Wood E (2006) Development of a 50-year
high-resolution global dataset of meteorological forcings for
land surface modeling. J Clim 19:3088–3111. doi:10.1175/
JCLI3790.1
Uppala SM, Kallberg PW, Simmons AJ, Andrae U, Da Costa
Bechtold V, Fiorino M, Gibson JK, Haseler J, Hernandez A,
Kelly GA, Li X, Onogi K, Saarinen S, Sokka N, Allan RP,
Andersson E, Arpe K, Balmaseda MA, Beljaars ACM, Van De
Berg L, Bidlot J, Bormann N, Caires S, Chevallier F, Dethof A,
Dragosavac M, Fisher M, Fuentes M, Hagemann S, Ho
´lm E,
Hoskins BJ, Isaksen L, Janssen PAEM, Jenne R, McNally AP,
Mahfouf J-F, Morcrette J-J, Rayner NA, Saunders RW, Simon P,
Sterl A, Trenberth KE, Untch A, Vasiljevic D, Viterbo P,
Woollen J (2005) The ERA-40 re-analysis. Q J R Metab Soc
131:2961–3012. doi:10.1256/qj.04.176
Vautard R and co-authors (2007) Summertime European heat and
drought waves induced by wintertime Mediterranean rainfall
deficit. Geophys Res Lett 34:L07711. doi:10.1029/2006
GL028001
Von Storch H, Zwiers F (1999) Statistical analysis in climate
research. Cambridge University Press, Cambridge
Watanabe M, Nitta T (1998) Relative impacts of snow and sea surface
temperature anomalies on an extreme phase in the winter
atmospheric circulation. J Clim 11:2837–2857. doi:10.1175/
1520-0442(1998)011\2837:RIOSAS[2.0.CO;2
WCRP (2008) WCRP Position paper on seasonal prediction. Report
from the 1st WCRP Seasonal Prediction workshop, Barcelona,
Spain, 4–7 June 2007. ICPO Publication, 127, 23p
818 H. Douville: Relative contribution of soil moisture and snow mass to seasonal climate predictability
123
