1086
JOURNAL OF HYDROMETEOROLOGY
VOLUME 12
Impact of Sea Surface Temperature and Soil Moisture on Summer Precipitation
in the United States Based on Observational Data
RUI MEI AND GUILING WANG
Department of Civil and Environmental Engineering, and Center for Environmental Sciences and Engineering,
University of Connecticut, Storrs, Connecticut
(Manuscript received 29 April 2010, in final form 4 February 2011)
ABSTRACT
This study examines the impact of sea surface temperature (SST) and soil moisture on summer precipitation
over two regions of the United States (the upper Mississippi River basin and the Great Plains) based on data
from observation and observation-forced model simulations (in the case of soil moisture). Results from SST–
precipitation correlation analysis show that spatially averaged SST of identified oceanic areas are better
predictors than derived SST patterns from the EOF analysis and that both predictors are strongly associated
with the Pacific Ocean. Results from conditioned soil moisture–precipitation correlation analysis show that
the impact of soil moisture on precipitation differs between the outer-quartiles years (with summer precipitation amount in the first and fourth quartiles) and inner-quartiles years (with summer precipitation
amount in the second and third quartiles), and also between the high- and low-skill SST years (categorized
according to the skill of SST-based precipitation prediction). Specifically, the soil moisture–precipitation
feedback is more likely to be positive and significant in the outer-quartiles years and in the years when the skill
of precipitation prediction based on SST alone is low. This study indicates that soil moisture should be included as a useful predictor in precipitation prediction, and the resulting improvement in prediction skills will
be especially substantial during years of large precipitation anomalies. It also demonstrates the complexity of
the impact of SST and soil moisture on precipitation, and underlines the important complementary roles both
SST and soil moisture play in determining precipitation.
1. Introduction
The variability of sea surface temperature (SST) affects variations of precipitation over land at seasonal to
interannual time scales (e.g., Hu and Feng 2001) through
its impact on atmospheric circulation, atmospheric energy, and moisture content. Because of the ocean’s large
heat capacity and thermal inertia, SST anomalies (SSTa)
tend to persist for a long time, making them useful predictors for regional climate—for precipitation in particular. Numerous studies have investigated the relationship
between SSTa and precipitation in North America. Precipitation over the continental United States is found to be
significantly associated with SSTa in the tropical and North
Pacific and in the Atlantic (Namias 1991; Trenberth and
Guillemot 1996; Hu and Feng 2001; Schubert et al.
2004a,b, 2008, 2009; Seager et al. 2005, 2008; Seager
Corresponding author address: Dr. Guiling Wang, Department of
Civil and Environmental Engineering, University of Connecticut,
261 Glenbrook Rd., Storrs, CT 06269-2037.
E-mail: [email protected]
DOI: 10.1175/2011JHM1312.1
Ó 2011 American Meteorological Society
2007; C. Wang et al. 2007; Weaver and Nigam 2008; Weaver
et al. 2009a,b). The physical mechanisms underlying the
teleconnections have also been explored in previous studies. For example, Seager et al. (2005) found that tropical
Pacific SSTa cause persistent drought or wet conditions
over the western United States through changes in the
subtropical jets, transient eddies, and the eddy-driven
mean meridional circulations. Schubert et al. (2004b,
2008) presented evidence that Pacific SSTa can affect
precipitation in Great Plains through El Niño–Southern
Oscillation (ENSO) impact on planetary waves (and
therefore the Pacific storm track), while Atlantic SSTa
influence precipitation through changes in the Bermuda
high- and low-level moisture flux into the continental
United States. The Great Plains low-level jet (GPLLJ)
is another suggested mechanism linking SSTa from surrounding oceans to regional hydroclimate of the central
United States (Weaver and Nigam 2008; Weaver et al.
2009a,b).
In addition to remote oceanic forcing, land surface
conditions provide an important local forcing for precipitation variability at subseasonal to seasonal time
OCTOBER 2011
MEI AND WANG
scales through water, energy, and momentum flux exchanges with the atmosphere. Soil moisture depletion
through evapotranspiration is an inherently slow process,
with the characteristic time scales ranging from weeks to
months or longer. Potential positive feedback between soil
moisture and precipitation, which tends to perpetuate and
sustain anomalous hydrological conditions such as floods
or droughts, promotes a long land memory and improves
predictability of the land–atmosphere system, leading to
the contribution of realistic soil moisture initialization
to subseasonal precipitation prediction (Dirmeyer 2000;
Koster and Suarez 2001; Dirmeyer et al. 2009; Koster et al.
2010). Soil moisture therefore may serve as a potential
predictor for precipitation over regions with a long land
memory and strong land–atmosphere coupling.
Theoretical analysis supports the notion of a positive soil moisture–precipitation feedback (Eltahir 1998;
Findell and Eltahir 2003a,b). A large number of numerical modeling studies have demonstrated this feedback over various regions of the United States including
the Midwest and the Great Plains (Bosilovich and Sun
1999; Pal and Eltahir 2001, 2002; Oglesby et al. 2002;
Koster et al. 2004, 2006, 2010; Kim and Wang 2007;
G. Wang et al. 2007; Dirmeyer et al. 2009). For example,
Kim and Wang (2007), using the coupled Community
Atmosphere Model, version 3 (CAM3)–Community Land
Model, version 3 (CLM3) model, found that soil moisture anomalies could affect summer precipitation over
the Mississippi River basin (Midwest), with the impact
depending on the characteristics of such anomalies, including their timing, magnitude, spatial coverage, and vertical
depth. Koster et al. (2004) identified the Great Plains
as a major region of strong soil moisture–precipitation
coupling in the Global Land–Atmosphere Coupling
Experiment 1 (GLACE1) based on 12 general circulation models (GCMs). Other studies found that the
strength of soil moisture impact in the model may be
influenced by model parameterizations related to surface water budget (Wu and Dickinson 2005; G. Wang
et al. 2007). In the follow-up experiment (GLACE2)
with a multimodel approach, Koster et al. (2010) found
moderate contribution of land surface initialization to
subseasonal precipitation prediction over limited areas
in the United States including part of the Great Plains.
Dirmeyer et al. (2009) applied different atmospheric reanalysis data to land surface models to assess the season
dependence of land–atmosphere feedback and found that
such feedback is strong through most of the year in the
Great Plains region.
In contrast to the abundance of evidence from numerical modeling studies, observational studies on
soil moisture–precipitation feedback are inconclusive
(Findell and Eltahir 1997; Salvucci et al. 2002; D’Odorico
1087
and Porporato 2004; Ruiz-Barradas and Nigam 2005,
2006; Zhang et al. 2008). For example, the study of
Findell and Eltahir (1997), an observation-based analysis
supporting the positive feedback between soil moisture
and precipitation in the U.S. Midwest, was questioned
by Salvucci et al. (2002) citing improper filtering of the
raw data. D’Odorico and Porporato (2004), using the
same data as Findell and Eltahir (1997), found that soil
moisture appears to influence the frequency of subsequent precipitation, but not the amount. Studies based
on reanalysis data (Ruiz-Barradas and Nigam 2005, 2006)
found very weak local precipitation recycling in the Great
Plains compared to model simulations, and suggested
that coupling strength in numerical models might have
been overestimated. Zhang et al. (2008) used soil moisture from the Global Land Data Assimilation System
(GLDAS) and observational precipitation to assess the
global land–atmosphere coupling, and found the northern continental United States to be one of the regions of
strong coupling.
It is clear based on previous studies that both oceanic
forcing and local forcing (soil moisture in particular)
influence precipitation over the continental United States.
However, most previous studies dealt with them separately. To improve the skill of the seasonal or subseasonal operational precipitation forecast, both factors
should be considered. How to systematically synthesize
the impact of the two, however, remains a challenge.
Very few observational studies have tackled this issue.
Hu and Feng (2004b) found that the effect of ‘‘land
surface processes’’ on summer precipitation over the
Southwest is strong, as indicated by a strong correlation
between antecedent winter and summer precipitation
when the persistency of influential summer SSTa in the
Pacific collapses, while the impact of summer SSTa is
strong when it persists. However, in their further study,
land surface processes are considered to be linked to soil
temperature rather than soil moisture (Hu and Feng
2004a). Similarly, Meng and Quiring (2009) found that
spring soil moisture is a good predictor of summer precipitation over the Great Plains region when Niño summer SSTa is not persistent, while Niño summer SSTa
is a good predictor when it is persistent. In addition, Wu
and Kinter (2009) carried out a correlation analysis with
SST, soil moisture, and observational drought indices
including both Palmer drought severity index (PDSI)
and standard precipitation index (SPI) of different time
scales, and found that the contribution of soil moisture
relative to that of SSTa is more significant in long-term
droughts (more than 6 months) than in short-term droughts
(less than 3 months) over the United States.
This study attempts to identify the statistical relationship between SST and seasonal and subseasonal
1088
JOURNAL OF HYDROMETEOROLOGY
precipitation for regions of North America and explores
how the impact of local soil moisture on precipitation
may depend on the impact of SST and on specific precipitation regimes. Section 2 briefly describes the data
used and its preprocessing. Methodology and results for
the SST–precipitation relationship and for the conditioned soil moisture–precipitation relationship are documented in sections 3 and 4, respectively. Section 5
presents the summary and discussions.
2. Data description and preprocessing
This study analyzes the impact of global SST and local
soil moisture on precipitation based on observational
data, focusing on two regions of the continental United
States: the Great Plains (defined as 37.58–458N, 1058–
958W) and the upper Mississippi River basin (defined as
368–448N, 928–858W). The definition of those two regions
is adopted from previous studies (Koster et al. 2004; Kim
and Wang 2007; Meng and Quiring 2009). The observational data used include global SST and precipitation and
soil moisture data over the United States.
The SST data are from the Met Office Hadley Centre
Sea Ice and Sea Surface Temperature, version 1.1
(HadISST1.1), a monthly global dataset spanning the period 1870–present, with a 18 3 18 spatial resolution, provided by the Met Office Hadley Centre (Rayner et al.
2003). In situ sea surface observations and satellite-derived
estimates at the sea surface are included in constructing
the dataset. SST bucket corrections have been applied
to gridded fields from 1870 through 1941, and a blend
of satellite Advanced Very High Resolution Radiometer
(AVHRR) data and in situ observations are used in the
modern periods.
The precipitation data used are the daily Climate Prediction Center (CPC) U.S. Unified Precipitation Dataset,
spanning the period 1948–97. The data are provided by
the National Oceanic and Atmospheric Administration
(NOAA)/Office of Oceanic and Atmospheric Research
(OAR)/Earth System Research Laboratory (ESRL)
Physical Sciences Division (PSD), Boulder, Colorado,
from their Web site at http://www.esrl.noaa.gov/psd/,
and cover the land area within 208–608N, 1408–608W
with a spatial resolution of 0.258 3 0.258. The data are
developed from multiple sources of rain gauge data and
have passed through standard quality controls conducted
by the CPC. Further details of this dataset can be found in
Higgins et al. (2000).
Daily soil moisture data, at 0.1258 3 0.1258 spatial resolution over the domain of 258–528N, 1248–678W, is derived from the simulation by the Variable Infiltration
Capacity (VIC) model for the United States from 1950 to
2000 (Maurer et al. 2002). The VIC model, a 1D macroscale
VOLUME 12
distributed hydrological model, developed at the University of Washington and Princeton University (Liang et al.
1996a,b), was driven with observed meteorological forcing
to produce the soil moisture data used here. The VIC daily
soil moisture includes data of three soil layers with depth
of each layer specified for each grid cell as derived during
the calibration process. Evaluation of simulated VIC soil
moisture against observation from the Soil Climate Analysis Network (SCAN) over several U.S. sites (Meng and
Quiring 2008) shows that VIC generally performs well in
simulating the soil moisture condition.
The analysis in this study will focus on the overlapping
period of all three datasets described above—that is,
1950–97. Prior to analysis, raw data have been preprocessed
as follows. Summer SSTa, referred to as SST below, are
obtained through averaging SST monthly anomalies
among June, July, and August at 18 3 18 grid spacing.
Daily soil moisture normalized anomalies, referred to
as daily soil moisture below, are derived for any single
layer or combination of multiple layers at 0.1258 3
0.1258 grid spacing. For example, the normalization for 1
June soil moisture is done by first subtracting the mean
soil moisture for 1 June and then dividing the anomalies
by the standard deviation of 1 June soil moisture. Daily
precipitation data are aggregated into several different
temporal resolutions—including the summer seasonal
total, monthly, and 21 day—at 0.258 3 0.258 grid spacing
to suit different needs of analysis. Summer total (sum of
June, July, and August) and monthly (June, July, and
August, respectively) precipitation normalized anomalies, referred to as summer and monthly precipitation
respectively below, are derived to pair with summer
SSTa for SST–precipitation relationship analysis. The
normalized anomalies of consecutive 21-day precipitation
(i.e., there are 92 data points in each summer; the first
point is the sum of precipitation over 2–22 June, the second over 3–23 June, and the last over 1–21 September),
referred to as 21-day precipitation (amount) below, are
derived to pair with the corresponding antecedent daily
soil moisture normalized anomalies (i.e., the first point is
1 June, and the last point is 31 August) in summer for
conditioned soil moisture–precipitation relationship analysis. Similarly, the normalization for 21-day precipitation
(e.g., 2–22 June) is done by first subtracting the mean
21-day precipitation and then dividing the anomalies by
the standard deviation of 21-day precipitation. In addition, precipitation frequencies in each 21-day period (i.e.,
number of rainy days during the 21-day period divided
by 21 days) at 0.258 3 0.258 grid spacing, referred to as
21-day precipitation frequency below (in contrast to 21-day
precipitation amount), are also derived to pair with antecedent daily soil moisture normalized anomalies in summer for comparison analysis. The normalized anomalies
OCTOBER 2011
MEI AND WANG
of precipitation and soil moisture are used because the
raw temporal averages and standard deviation of those
two hydrological variables can vary greatly across grids
(over the Great Plains or the upper Mississippi River basin)
or months (because of intraseasonal variability). Normalized anomalies can represent the degree of dryness or
wetness across the grids or months in a comparable way.
That is similar to the standard precipitation index (McKee
et al. 1993).
3. SST–precipitation relationship
a. Methodology
Two different approaches to analyzing the SST–
precipitation correlation are examined. One is to correlate
summer precipitation averaged for a specific region (i.e.,
the upper Mississippi River basin or the Great Plains) with
summer SST of the globe. This will produce a global map
of correlation, which can be used to identify oceanic areas
with the most significant signal of teleconnection. The
time series of summer SST averaged over those areas can
then be correlated with summer precipitation to develop
a SST–precipitation relationship for the upper Mississippi
River basin and the Great Plains, respectively.
The other approach is based on the empirical orthogonal function (EOF) analysis. EOF and rotated EOF
(REOF; Richman 1986) methods have been explored to
derive SST modes in many previous studies (Zhang et al.
1997; Barlow et al. 2001; Castro et al. 2007; Schubert
et al. 2002, 2009). These studies have shown that the
regular EOF analysis is very sensitive to SST domains and
the REOF analysis (mostly over the globe) works better
than the regular EOF to extract different physical modes
separately. In this study we isolate the leading patterns of
the interannual variability of summer SST using both the
EOF and REOF methods, and identify the EOF patterns
(EOFs) and REOF patterns (REOFs) with significant
correlations to summer precipitation for the two regions
of interest. The time series of the EOFs and REOFs can
then be correlated with the time series of summer precipitation to develop a SST–precipitation relationship.
Note that although the SST–precipitation relationship
analysis below is based on a simultaneous correlation
instead of a lagged correlation, it may still be used for
prediction purpose as robust SST forecasts are routinely
available (e.g., from the International Research Institute
for Climate and Society at http://iri.columbia.edu/climate/
forecast/sst/).
b. Results on SST–precipitation relationship
In the first approach, summer and monthly (June, July,
and August) precipitation averaged for the upper Mississippi River basin and the Great Plains are correlated
1089
with gridded SST to produce the corresponding global
correlation maps (Fig. 1). While there are scattered areas
of significant correlation around the global ocean, most of
the areas that show temporally persistent correlation are
located in the Pacific Ocean, suggesting that the Pacific
Ocean plays an important role in influencing precipitation
in both the upper Mississippi River basin and the Great
Plains. From Fig. 1 in the Pacific, SST area 1 (308–358N,
1808–1658W) is identified as an area of strong teleconnection for precipitation over the upper Mississippi River
basin, and SST areas 1 and 2 (1628–1528W, 08–88N) are
areas of strong teleconnection for precipitation over the
Great Plains. Comparison with the spatial pattern of
known climate modes (Zhang et al. 1997) indicates that
area 1 is located near the Pacific decadal oscillation
(PDO) center of action and area 2 is located near the
ENSO center of action. The strong correlations found in
these two areas may be partly related to the impact of
these known climate modes on precipitation over the
continental United States, but they explain more precipitation variance than the ENSO and PDO indices do.
In the other approach, EOFs of summer SST over
1950–97 are derived for three domains: Pacific (208S–
608N, 1208E–1008W), Atlantic (208–608N, 1008W–308E),
and global (608S–608N, 1808W–1808E), and REOFs are
derived only for the global domain as in Schubert et al.
(2009). EOFs that significantly correlate with precipitation over the Great Plains are presented in Fig. 2,
including Pacific EOF1, Pacific EOF2, global EOF1, and
global EOF3. Both the Pacific EOF1 and the global
EOF1 are an ENSO-like pattern with an out-of-phase
relationship between the central and eastern equatorial
Pacific and the central North Pacific, which also resemble the ‘‘Pacific forcing pattern’’ in Schubert et al.
(2009). Their associated principal components (PCs) are
highly correlated (R2 5 0.96), and both are also highly
correlated with ENSO and PDO, and include both interannual and interdecadal signals. Note that the two
identified SST areas based on the first approach are located near the centers of action of the Pacific or global
EOF1. In addition, the Atlantic EOF1 resembles the
Atlantic multidecadal oscillation pattern (Enfield et al.
2001) with moderate correlation (R2 5 0.43); the REOFs
over the globe resemble those mentioned physical modes
to a much less extent than the regular EOFs.
To further quantify the relationship between SST and
precipitation, linear regressions using the leave-one-out
cross-validation method (i.e., prediction in any specific
year is based on the regression equation derived using
the other 47-year data points) are conducted, linking
summer precipitation in both regions to the EOFs and to
the spatially averaged SSTs, respectively (Fig. 3). In the
upper Mississippi River basin, only spatially averaged
1090
JOURNAL OF HYDROMETEOROLOGY
VOLUME 12
FIG. 1. The global correlation map between spatially averaged precipitation (summer, June, July, and August) of each region
and gridded global SST. The delineated SST area 1 (308–358N, 1808–1658W) for the upper Mississippi River basin and SST areas 1 and 2
(08–88N, 1628–1528W) for the Great Plains are identified as areas of strong teleconnection and marked with black solid lines.
SST can provide significant precipitation prediction at
95% confidence level, which is based on SST in area 1;
the SSTs can explain 37% of total precipitation variances.
Therefore, only precipitation predicted by SST in area 1 is
plotted against observation in the top panel of Fig. 3. For
the Great Plains region, both EOFs and area-averaged
SST can provide significant prediction at 95% confidence
level, but the prediction based on area-averaged SSTs is
better than the prediction based on the Pacific EOF1 and 2
(Fig. 3) and better than any other combination of EOFs
from all the domains examined. The area-averaged SST,
the Pacific EOFs, and the global EOFs can explain 47%,
28%, and 23% of total precipitation variances, respectively. It is clear from Fig. 3 that the statistical relationship
between SST and precipitation in the Great Plains is
stronger than in the upper Mississippi River basin.
4. Conditioned soil moisture–precipitation
relationship
a. Methodology
Previous studies demonstrated that the impact of soil
moisture anomalies on precipitation may be stronger
when soil moisture anomalies are larger (e.g., Oglesby
et al. 2002; Koster et al. 2010) or during the periods when
SST is a poor predictor for precipitation (e.g., Hu and
Feng 2004b; Meng and Quiring 2009). Therefore, the
soil moisture–precipitation relationship is analyzed using
a two-step categorized correlation analysis approach—that
is, a conditioned correlation analysis. All 48 years are first
divided into different categories; correlation is then analyzed for each category.
Two different methods of categorization are examined.
In the first method, all the 48 years of data are divided
into two categories, with equal number of years in each
(24 years), based on the amount of summer precipitation:
one outer-quartiles category (i.e., too much or too little
precipitation) and one inner-quartiles category. Summer
precipitation for each region of interest (the Great Plains
or the upper Mississippi River basin) is used as the criterion to rank the years based on its amount: the first and
fourth quartiles are grouped in the outer-quartiles category, and the second and third quartiles are grouped in
the inner-quartiles category.
In the other method, the data are divided into two
categories with equal number of years (24 years) according to the skill of SST in predicting precipitation:
one category with precipitation well predicted by SST
OCTOBER 2011
MEI AND WANG
1091
FIG. 2. (left) EOF1 and 2 of Pacific summer SST and (right) EOF1 and 3 of global summer SST for the period
1950–97. They are presented as the homogeneous correlation between time series of EOF pattern and time series of
gridded SST (value above each panel indicates the explained percentage of total variances).
forcing (high-skill SST) and the other poorly predicted
by SST forcing (low-skill SST). The absolute error
between observed summer precipitation and the estimate from cross-validation analysis based on SST area
averages is used as the criteria to rank the years: the
upper half are grouped in the low-skill SST category and
the lower half are grouped in the high-skill SST category.
For each of the two different categorizations, two different approaches are used to investigate the soil moisture–
precipitation correlation. One is to analyze the evolution of
FIG. 3. Observed and SST-predicted time series of summer precipitation during the period
1950–97 for each region.
1092
JOURNAL OF HYDROMETEOROLOGY
VOLUME 12
FIG. 4. Correlation between 1-day soil moisture and precipitation amount (star solid lines) or
precipitation frequency (square solid lines) in the subsequent 21 days for each region. Here 5%
significance lines for the sample size of 48 derived from a Monte Carlo simulation are in solid
and dashed lines for amount and frequency, respectively.
temporal correlation through the whole summer between
1-day soil moisture and precipitation amount or frequency
in the subsequent 21-day period with no temporal gap in
between (i.e., soil moisture leading by 1 day), as in Findell
and Eltahir (1997) and D’Odorico and Porporato (2004),
within each category. For example, the correlation on
1 June for a specific category is derived based on daily soil
moisture on 1 June and precipitation amount or frequency
over the period 2–22 June for each of the 24 years. To
represent the average level of correlation with sample size
of 24 and for comparison with categorized correlation for
each category, the referenced correlation is derived following the procedure below using 1 June as an example:
1) 24 pairs of daily soil moisture data on 1 June and 21-day
precipitation amount (or frequency) during 2–22 June are
drawn without replacement from the 48 total samples, and
the correlation is computed based on these pairs of data.
2) The first procedure is repeated 10 000 times to derive
the probability distribution function (PDF) of correlation.
3) The corresponding 50th percentile correlation value is
extracted from this PDF as the referenced correlation.
Namely, the correlation on 1 June derived for each different category is simply one realization from those forming
the PDF for 1 June. In addition, the temporal evolution of
correlation with 95% confidence level is derived for significance test using different sample sizes, with 48 for the
whole group of data and 24 for each categorization, based
on a Monte Carlo simulation (Wilks 1995, 145–157).
The other approach is to lump the data together and
include not only interannual but also intraseasonal
variability in the correlation analysis, and compare the
statistics of soil moisture–precipitation correlation among
different categories. In each category, the total number
of data pairs is 24 3 92, where 24 is the number of years
in each category and 92 is the number of days in the
summer. To estimate one correlation value, 24 data pairs
are randomly drawn without replacement from the pool
of size 24 3 92 within each category (or 48 3 92 for the
whole group) with a constraint to exclude temporal overlapping between any two data points of precipitation. For
example, in the calculation of one correlation value, if
the pair of data with soil moisture on 6 June and precipitation during 7–27 June in some year is drawn, the
closest neighboring data pair that can potentially be
drawn is soil moisture on 27 June and precipitation during
28 June–18 July in the same year. The corresponding
correlation between 1-day soil moisture and subsequent
21-day precipitation amount or frequency is calculated
based on those 24 data pairs, and this procedure is repeated 10 000 times to obtain the PDF of the correlation
under each categorization. Note that both soil moisture
and precipitation (amount or frequency) used for correlation analysis are spatial averages over each of the two
regions of interest.
b. Results on soil moisture–precipitation correlation
Depending on the depth of plant rooting zone, soil
moisture at different depths may differ in their impact
on atmospheric processes. To identify the soil layer with
the strongest impact on precipitation, we examine the
OCTOBER 2011
MEI AND WANG
1093
FIG. 5. Temporal correlation between (a),(b) 1-day soil moisture and precipitation amount or (c),(d) frequency in
the subsequent 21 days based on the two categories: outer-quartiles (square solid lines) and inner-quartiles (star solid
lines) categorized according to the amount of summer precipitation for each region with the corresponding referenced 50th percentile correlation in solid lines. Here 5% significance lines for the sample size of 24 derived from
a Monte Carlo simulation for the correlation are dashed.
evolution of correlation between 1-day soil moisture
using different soil layers and precipitation amount in
the subsequent 21 days through most of the year based
on the whole data record over the period of 1950–97 (not
shown here). During summer, the top layer soil moisture
is generally better correlated with precipitation than any
other single layer or multiple layers combined. The top
layer soil moisture is therefore used for all subsequent
correlation analysis.
Correlation of course is not always a result of causal
relationship. Specifically, a positive soil moisture–
precipitation correlation can potentially result from
precipitation persistence not attributed to soil moisture–
precipitation feedback. However, in the two continental
U.S. regions analyzed here, precipitation autocorrelation
(i.e., correlation between adjacent 21-day precipitations,
not shown here) is weaker than soil moisture–precipitation
correlation, suggesting that the soil moisture–precipitation
correlation found here is a strong indication of soil
moisture–precipitation feedback and not an artificial
effect of precipitation persistence.
Figure 4 illustrates the correlation between 1-day soil
moisture and precipitation amount versus precipitation
frequency in the subsequent 21 days in summer for both
regions. The results show that most of the time the correlation between soil moisture and subsequent precipitation frequency is better than that between soil moisture
and subsequent precipitation amount, which agrees with
the results in D’Odorico and Porporato (2004) and indicates that soil moisture–precipitation feedback is more
likely to take place through effects on precipitation frequency rather than amount. Both precipitation amount
and frequency are considered in our following analysis.
Figure 4 also shows that correlation between soil moisture and precipitation in the Great Plains is better than
that in the upper Mississippi River basin, indicating a
stronger land–atmosphere coupling in the Great Plains
than in the upper Mississippi River basin. This result is
therefore consistent with the stronger coupling strength
found in the Great Plains in numerical modeling studies
(e.g., Koster et al. 2004, 2006).
Temporal correlations through summer between 1-day
soil moisture and precipitation amount or frequency in
the subsequent 21 day conditioned on the magnitude
of summer precipitation anomalies together with the
referenced 50th percentile correlation are presented in
Fig. 5 for both regions. Evidently in Figs. 5a,b, correlation between soil moisture and precipitation amount in
1094
JOURNAL OF HYDROMETEOROLOGY
VOLUME 12
FIG. 6. Temporal correlation between (a),(b) 1-day soil moisture and precipitation amount or (c),(d) frequency in
the subsequent 21 days based on the two categories: high-skill SST (star solid lines) and low-skill SST (square solid
lines) categorized according to the skill of SST in predicting precipitation for each region with the corresponding
referenced 50th percentile correlation in solid lines. Here 5% significance lines for the sample size of 24 derived from
a Monte Carlo simulation for the correlation are dashed.
the outer-quartiles category is stronger and more persistent than in the inner-quartiles category for both regions, with the former sitting above and the latter below
the referenced 50th percentile correlation. This implies
that in the years with large precipitation anomalies,
soil moisture–precipitation feedback (therefore coupling strength) is stronger. Note that in the Great Plains,
the correlation in the outer-quartiles category is much
stronger, indicating that soil moisture may lead to more
substantial improvement of precipitation prediction in
this region for the years with large precipitation anomalies. Figures 5c,d provide similar insight into the soil
moisture–precipitation feedback between these two categories using precipitation frequency rather than precipitation amount.
Figure 6 presents the temporal evolution of correlations between 1-day soil moisture and the subsequent
21-day precipitation amount or frequency for the two
categories of SST-based precipitation prediction (highand low-skill SST), compared with the referenced 50th
percentile correlation. Analogous to Fig. 5 and regardless of whether precipitation amount or frequency is used,
correlation between soil moisture and precipitation is
more persistent and positive in the low-skill SST category
than in the high-skill SST category, with the former staying above and the latter below the average level of correlation represented by the referenced 50th percentile
correlation. In the Great Plains, the correlation in the lowskill SST category is also stronger than that in the upper
Mississippi River basin. It suggests that soil moisture–
precipitation feedback is stronger during the years when
SST does not perform well in predicting precipitation, and
therefore soil moisture plays an important role complementary to SST in improving precipitation prediction.
Figures 7 and 8 show the PDF of correlation between
1-day soil moisture and precipitation amount in the subsequent 21 day, derived when data for all summer days
in all relevant years are lumped together, for the two
different categorization approaches, respectively. Note
that the correlations shown in Figs. 5 and 6 are samples
of these PDFs.
In the upper Mississippi River basin, the correlation
with peak probability density is positive in the outerquartiles category, slightly positive (around zero) without categorization, and negative in the inner-quartiles
category. In the Great Plains, the correlation with peak
probability density is all positive, strongest in the outerquartiles category, weaker when not categorized, and
OCTOBER 2011
MEI AND WANG
1095
FIG. 7. Probability distribution function of correlation between
1-day soil moisture and the subsequent 21-day precipitation, derived through sampling the 48-yr whole group and each of the two
24-yr subgroups (outer and inner quartiles) categorized according
to the amount of summer precipitation in each region. The sample
size for all correlation calculation is 24.
FIG. 8. Probability distribution function of correlation between
1-day soil moisture and the subsequent 21-day precipitation, derived through sampling the 48-yr whole group and each of the two
24-yr subgroups (high- and low-skill SST) categorized according
to the skill of SST in predicting precipitation in each region. The
sample size for all correlation calculation is 24.
close to zero in the inner-quartiles category (Fig. 7).
These provide further statistical confidence for results in
Fig. 5 and suggest that soil moisture–precipitation feedback is more likely to be positive and significant in the
years with large precipitation anomalies.
In the same way, Fig. 8 illustrates the PDF of correlation based on the uncategorized data and the two categories of different SST skills in precipitation prediction.
In the upper Mississippi River basin, the correlation with
peak probability density is positive in the low-skill SST
category and is around zero in the high-skill SST category; in the Great Plains, it is positive in both categories
and is much stronger in the low-skill SST category than in
the high-skill SST category. For both regions, the correlation with peak probability density derived from uncategorized data falls between the two categories. These
provide statistical confidence for results in Fig. 6 and
suggest that soil moisture–precipitation feedback is more
likely to be positive and significant during the years when
SST alone performs poorly in predicting precipitation.
Note that the PDFs of correlation between soil moisture
and precipitation frequency (not shown here) are consistent with the implications on soil moisture–precipitation
frequency feedback reflected in Figs. 5 and 6. Also note
that comparison between PDFs indicates that correlation
between soil moisture and precipitation frequency is
better than that between soil moisture and precipitation
amount, which is consistent with results in Fig. 4.
As a sensitivity experiment, the lagged correlation
between 1-day soil moisture and 21-day precipitation
with soil moisture leading by 3 days and 1 week has been
conducted for all the analyses above. The results present
no qualitative difference in terms of the comparison
between categorized correlations—that is, the correlation between soil moisture and precipitation is stronger
in the outer-quartiles years than in the inner-quartiles
years, and is stronger in low-skill SST years than in highskill SST years. However, as the time lag increases, the
correlation between soil moisture and precipitation becomes weaker. For example, Fig. 9 shows the PDFs of
1096
JOURNAL OF HYDROMETEOROLOGY
FIG. 9. Probability distribution function of correlation between
1-day soil moisture and future 21-day precipitation with 1-, 3-, and
7-day lags, derived based on the pool of the 24-yr outer-quartiles
category for the Great Plains. The sample size for all correlation
calculation is 24.
correlation between 1-day soil moisture and 21-day
precipitation with soil moisture leading by 1, 3, and 7
days for the outer-quartiles years in the Great Plains,
with the correlation coefficient of peak density changing
from 0.38 at 1-day lead to 0.28 at 1-week lead time. In
addition, the results (not shown here) of comparison
between categorized correlations still hold if soil water
in the top two or three layers combined is used instead of
only the top layer, or if a 14- or 28-day window is used
instead of 21 day for precipitation. Note that the 21-day
window is used because it accommodates a comparison
between our study and previous studies (Findell and
Eltahir 1997; D’Odorico and Porporato 2004), and it is
consistent with the time scale of our interest (i.e., seasonal
and subseasonal) as opposed to the weather time scale,
which is a few days. Complementary correlation analysis
(not shown here) between daily soil moisture and subsequent precipitation summed over different accumulation
periods (not shown here) indicates that the correlation is
stronger with relatively shorter precipitation accumulation
periods (e.g., 3 and 5 day) than with longer ones (e.g., 7, 14,
and 21 day). While the correlation with precipitation of
a longer accumulation period always contains contribution
from shorter periods, the analysis after removing precipitation in the immediate following days (Fig. 9) indicates
substantial contribution from the relationship between soil
moisture and precipitation further into the future.
From Figs. 5–8, for both categorization approaches,
comparison of soil moisture–precipitation correlations
between the two regions of interest in the United States
indicates that soil moisture, like SST, also has better
prediction skills in the Great Plains than in the upper
Mississippi River basin. Comparison between the years
VOLUME 12
in the outer-quartiles category (24 years) and those in
the low-skill SST category (24 years) shows that 14 out
of 24 years within the outer-quartiles category fall into
the low-skill SST category for both regions. This has two
important implications: prediction by SST alone is usually poor in the years with large precipitation anomalies,
and soil moisture is critical in order to accurately predict
large precipitation anomalies.
One potential concern over the conditioned soil
moisture–precipitation relationship is that the stronger
correlation in the outer-quartiles category may reflect
a noise because of the nature of more scattering of large
anomalies instead of a true signal arising from land–
atmosphere coupling. However, this should not be the
case because the categorization here is based on summer
precipitation instead of the 1-day soil moisture or 21-day
precipitation. To further illustrate the difference, a categorization based on 1-day soil moisture is considered as
an example. Figure 10 presents the PDFs of daily soil
moisture over the two regions under the two categorization
methods: one as used in our analysis categorized based on
total summer precipitation, and the other categorized directly based on daily soil moisture. With the categorization
applied in this study, while daily soil moisture does show
more scattering in the outer quartiles, no clear separation
of the two categories is found. With the categorization
based on daily soil moisture, not surprisingly, the outerquartiles category presents two peaks at the tails indicating
much more scattering. Despite the much stronger scattering of outer-quartiles category in the alternative categorization method, the correlation of the outer-quartiles
category based on the categorization used in the study is
still stronger (Fig. 11), indicating that such strong correlation might be a true signal of land–atmosphere coupling.
5. Summary and discussions
In this study, the impact of sea surface temperature
and soil moisture on precipitation in summer is investigated based on observational data focusing on two
regions of the United States: the upper Mississippi River
basin and the Great Plains. Important findings include
the following.
1) Compared with the derived SST patterns from the EOF
analysis, spatially averaged SST over some identified
oceanic areas are better predictors for summer precipitation in both U.S. regions. Predictors derived from
both approaches are strongly associated with the Pacific
Ocean, indicating its important role in influencing precipitation over the continental United States (Seager
et al. 2005; Schubert et al. 2004b, 2008, 2009).
2) The correlation between soil moisture and subsequent precipitation frequency for both U.S. regions
OCTOBER 2011
MEI AND WANG
1097
FIG. 10. Probability distribution function of daily soil moisture over the two regions under two categorizations
(both include outer and inner quartiles): one based on summer total precipitation and the other based on daily soil
moisture.
in summer is stronger than the correlation between
soil moisture and subsequent precipitation amount,
which is consistent with D’Odorico and Porporato
(2004) using another observational dataset.
3) For both U.S. regions and regardless of whether
precipitation amount or frequency is considered, the
conditioned soil moisture–precipitation correlation
is stronger during the years with large summer precipitation anomalies, and is stronger during years
when SST presents low skill in summer precipitation
prediction than during years when SST exhibits high
skill. There is a substantial overlap between outerquartiles years and low-skill SST years, highlighting
the critical importance of including soil moisture in
predicting climate extremes.
4) Compared with the upper Mississippi River basin,
the Great Plains region may benefit from a better
SST-based precipitation prediction and a stronger
soil moisture–precipitation correlation.
On one hand, our finding that soil moisture and SST
may play complementary roles in precipitation prediction over the Great Plains is conceptually consistent
with Hu and Feng (2004a,b) and Meng and Quiring
(2009). However, our study differs from these previous
studies in two aspects: 1) Hu and Feng (2004a,b) and
Meng and Quiring (2009) found that local forcing and
remote forcing (SST) are responsible for precipitation
variability during alternating epochs of decades, but
in our study the impact of those two forcings is examined for individual years; and 2) these previous
studies focused on seasonal total precipitation while
our study explores both interannual variability of summer precipitation and subseasonal variability within
summer.
On the other hand, our finding regarding soil moisture–
precipitation feedback, with respect to precipitation
amount, are consistent with Oglesby et al. (2002) and
Koster et al. (2010) in the sense that there is a threshold
effect on the impact of soil moisture on subsequent precipitation. Oglesby et al. (2002) found that the impact
of initial dry soil moisture anomalies imposed over the
Mississippi–Missouri River basin on subsequent local
precipitation is only significant when the magnitude of
such anomalies are larger than a certain threshold. More
recently, Koster et al. (2010) found that the precipitation
forecast skills attributed to realistic land surface initialization increase under conditions with larger initial soil
moisture anomalies over the United States. Although the
nature of all those findings is subject to further research,
the collective implications are 1) realistic initialization of
soil moisture conditions are important to improving the
skill of subseasonal precipitation forecasts using numerical models, and 2) without accurate initialization of soil
1098
JOURNAL OF HYDROMETEOROLOGY
FIG. 11. Probability distribution function of correlation between
1-day soil moisture and subsequent 21-day precipitation in the
outer-quartiles category when categorization is based on summer
total precipitation and daily soil moisture, respectively. The sample
size for all correlation calculation is 24.
moisture, the forecast skill will suffer the most during
years with large precipitation anomalies when the forecast matters the most.
Acknowledgments. This research is supported by funding from the NOAA CPPA (Grant NA08OAR4310871).
We thank the two anonymous reviewers for their
constructive comments on an earlier version of the
manuscript.
REFERENCES
Barlow, M., S. Nigam, and E. H. Berbery, 2001: ENSO, Pacific
decadal variability, and U.S. summer time precipitation,
drought, and stream flow. J. Climate, 14, 2105–2128.
Bosilovich, M. G., and W. Y. Sun, 1999: Numerical simulation of
the 1993 midwestern flood: Land–atmosphere interactions.
J. Climate, 12, 1490–1505.
Castro, C. L., R. A. Pielke Sr., and J. O. Adegoke, 2007: Investigation of the summer climate of the contiguous United
States and Mexico using the Regional Atmospheric Modeling
System (RAMS). Part I: Model climatology (1950–2002).
J. Climate, 20, 3844–3865.
VOLUME 12
D’Odorico, P., and A. Porporato, 2004: Preferential states in soil
moisture and climate dynamics. Proc. Natl. Acad. Sci. USA,
101, 8848–8851.
Dirmeyer, P. A., 2000: Using a global soil wetness dataset to improve seasonal climate simulation. J. Climate, 13, 2900–2922.
——, C. A. Schlosser, and K. L. Brubaker, 2009: Precipitation,
recycling, and land memory: An integrated analysis. J. Hydrometeor., 10, 278–288.
Eltahir, E. A. B., 1998: A soil moisture–rainfall feedback mechanism. 1. Theory and observations. Water Resour. Res., 34,
765–776.
Enfield, D., A. Mestas-Nunez, and P. Trimble, 2001: The Atlantic
Multidecadal Oscillation and its relation to rainfall and river
flows in the continental U.S. Geophys. Res. Lett., 28, 2077–2080.
Findell, K. L., and E. A. B. Eltahir, 1997: An analysis of the soil
moisture-rainfall feedback, based on direct observations from
Illinois. Water Resour. Res., 33, 725–735.
——, and ——, 2003a: Atmospheric controls on soil moisture–
boundary layer interactions. Part I: Framework development.
J. Hydrometeor., 4, 552–569.
——, and ——, 2003b: Atmospheric controls on soil moisture–
boundary layer interactions. Part II: Feedbacks within the
continental United States. J. Hydrometeor., 4, 570–572.
Higgins, R. W., W. Shi, E. Yarosh, and R. Joyce, 2000: Improved
United States Precipitation Quality Control System and Analysis. NCEP/Climate Prediction Center Atlas 7, 40 pp.
Hu, Q., and S. Feng, 2001: Variations of teleconnection of ENSO
and interannual variation in summer rainfall in the central
United States. J. Climate, 14, 2469–2480.
——, and ——, 2004a: A role of the soil enthalpy in land memory.
J. Climate, 17, 3633–3643.
——, and ——, 2004b: Why has the land memory changed?
J. Climate, 17, 3236–3243.
Kim, Y., and G. Wang, 2007: Impact of initial soil moisture anomalies
on subsequent precipitation over North America in the coupled
land–atmosphere model CAM3–CLM3. J. Hydrometeor., 8,
513–533.
Koster, R. D., and M. J. Suarez, 2001: Soil moisture memory in
climate models. J. Hydrometeor., 2, 558–570.
——, and Coauthors, 2004: Regions of strong coupling between soil
moisture and precipitation. Science, 305, 1138–1140.
——, and Coauthors, 2006: GLACE: The Global Land–Atmosphere
Coupling Experiment. Part 1: Overview. J. Hydrometeor., 7,
590–610.
——, and Coauthors, 2010: Contribution of land surface initialization to subseasonal forecast skill: First results from a multimodel experiment. Geophys. Res. Lett., 37, L02402, doi:10.1029/
2009GL041677.
Liang, X., D. P. Lettenmaier, and E. F. Wood, 1996a: One-dimensional
statistical dynamic representation of subgrid spatial variability
of precipitation in the two-layer variable infiltration capacity
model. J. Geophys. Res., 101 (D16), 21 403–21 422.
——, E. F. Wood, and D. P. Lettenmaier, 1996b: Surface soil
moisture parameterization of the VIC-2L model: Evaluation
and modifications. Global Planet. Change, 13, 195–206.
Maurer, E. P., A. W. Wood, J. C. Adam, D. P. Lettenmaier, and
B. Nijssen, 2002: A long-term hydrologically based dataset of
land surface fluxes and states for the conterminous United
States. J. Climate, 15, 3237–3251.
McKee, T. B., N. J. Doesken, and J. Kleist, 1993: The relationship
of drought frequency and duration to time scales. Preprints,
Eighth Conf. on Applied Climatology, Anaheim, CA, Amer.
Meteor. Soc., 179–186.
OCTOBER 2011
MEI AND WANG
Meng, L., and S. M. Quiring, 2008: A comparison of soil moisture
models using Soil Climate Analysis Network observations.
J. Hydrometeor., 9, 641–659.
——, and ——, 2009: Observational relationship of sea surface
temperature and precedent soil moisture with summer precipitation in the U.S. Great Plains. Int. J. Climatol., 30, 884–
893, doi:10.1002/joc.1941.
Namias, J., 1991: Spring and summer 1988 drought over the contiguous United States—Causes and prediction. J. Climate, 4, 54–65.
Oglesby, R. J., S. Marshall, D. J. Erickson III, J. O. Roads, and F. R.
Robertson, 2002: Thresholds in atmosphere–soil moisture interactions: Results from climate model studies. J. Geophys.
Res., 107, 4244, doi:10.1029/2001JD001045.
Pal, J. S., and E. A. B. Eltahir, 2001: Pathways relating soil moisture
conditions to future summer precipitation within a model of
the land–atmosphere system. J. Climate, 14, 1227–1242.
——, and ——, 2002: Teleconnections of soil moisture and rainfall
during the 1993 midwest summer flood. Geophys. Res. Lett.,
29, 1865, doi:10.1029/2002GL014815.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V.
Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003:
Global analyses of sea surface temperature, sea ice, and night
marine air temperature since the late nineteenth century.
J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670.
Richman, M. B., 1986: Rotation of principal components. J. Climatol., 6, 293–335.
Ruiz-Barradas, A., and S. Nigam, 2005: Warm season rainfall
variability over the U.S. Great Plains in observations, NCEP
and ERA-40 reanalyses, and NCAR and NASA atmospheric
model simulations. J. Climate, 18, 1808–1830.
——, and ——, 2006: Great Plains hydroclimate variability: The
view from North American Regional Reanalysis. J. Climate,
19, 3004–3010.
Salvucci, G. D., J. A. Saleem, and R. Kaufmann, 2002: Investigating
soil moisture feedbacks on precipitation with tests of Granger
causality. Adv. Water Resour., 25, 1305–1312.
Schubert, S. D., M. J. Suarez, P. J. Pegion, M. A. Kistler, and
A. Kumer, 2002: Predictability of zonal means during boreal
summer. J. Climate, 15, 420–434.
——, ——, ——, R. D. Koster, and J. T. Bacmeister, 2004a: Causes
of long-term drought in the U.S. Great Plains. J. Climate, 17,
485–503.
——, ——, ——, ——, and ——, 2004b: On the cause of the 1930s
dust bowl. Science, 303, 1855–1859.
——, ——, ——, ——, and ——, 2008: Potential predictability of
long-term drought and pluvial conditions in the U.S. Great
Plains. J. Climate, 21, 2919–2937.
——, and Coauthors, 2009: A U.S. CLIVAR project to assess
and compare the responses of global climate models to
1099
drought-related SST forcing patterns: Overview and results.
J. Climate, 22, 5251–5272.
Seager, R., 2007: The turn of the century North American drought:
Global context, dynamics, and past analogs. J. Climate, 20,
5527–5552.
——, Y. Kushnir, C. Herweijer, M. Naik, and J. Velez, 2005:
Modeling of tropical forcing of persistent droughts and pluvials over western North America: 1856–2000. J. Climate, 18,
4065–4088.
——, ——, M. F. Ting, M. Cane, N. Naik, and J. Miller, 2008:
Would advance knowledge of 1930s SSTs have allowed prediction of the dust bowl drought? J. Climate, 21, 3261–3280.
Trenberth, K. E., and C. J. Guillemot, 1996: Physical processes
involved in the 1988 drought and 1993 floods in North
America. J. Climate, 9, 1288–1298.
Wang, C., S. K. Lee, and D. B. Enfield, 2007: Impact of the Atlantic
warm pool on the summer climate of the Western Hemisphere. J. Climate, 20, 5021–5040.
Wang, G., Y. Kim, and D. Wang, 2007: Quantifying the strength of
soil moisture–precipitation coupling and its sensitivity to
changes in surface water budget. J. Hydrometeor., 8, 551–570.
Weaver, S. J., and S. Nigam, 2008: Variability of the Great Plains
low-level jet: Large-scale circulation context and hydroclimate
impacts. J. Climate, 21, 1532–1551.
——, A. Ruiz-Barradas, and S. Nigam, 2009a: Pentad evolution of
the 1988 drought and 1993 flood over the Great Plains: An
NARR perspective on the atmospheric and terrestrial water
balance. J. Climate, 22, 5366–5384.
——, S. Schubert, and H. L. Wang, 2009b: Warm season variations
in the low-level circulation and precipitation over the central
United States in observations, AMIP simulations, and idealized SST experiments. J. Climate, 22, 5401–5420.
Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences.
International Geophysics Series, Vol. 59, Academic Press,
467 pp.
Wu, R. G., and J. L. Kinter III, 2009: An analysis of the relationship
of U.S. droughts with SST and soil moisture: Distinguishing
the time scale of droughts. J. Climate, 22, 4520–4538.
Wu, W., and R. E. Dickinson, 2005: Warm-season rainfall variability over the U.S. Great Plains and its correlation with
evapotranspiration in a climate simulation. Geophys. Res.
Lett., 32, L17402, doi:10.1029/2005GL023422.
Zhang, J., W.-C. Wang, and J. Wei, 2008: Assessing landatmosphere coupling using soil moisture from the Global
Land Data Assimilation System and observational precipitation. J. Geophys. Res., 113, D17119, doi:10.1029/
2008JD009807.
Zhang, Y., J. M. Wallace, and D. Battisti, 1997: ENSO-like interdecadal variability: 1900–93. J. Climate, 10, 1004–1020.