Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Brocca Et Al-2014-Journal of Geophysical Research Atmospheres

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

PUBLICATIONS

Journal of Geophysical Research: Atmospheres


RESEARCH ARTICLE Soil as a natural rain gauge: Estimating global
10.1002/2014JD021489
rainfall from satellite soil moisture data
Key Points: Luca Brocca1, Luca Ciabatta1, Christian Massari1, Tommaso Moramarco1, Sebastian Hahn2,
• New method to measure terrestrial
rainfall from satellite soil moisture data Stefan Hasenauer2, Richard Kidd2, Wouter Dorigo2, Wolfgang Wagner2, and Vincenzo Levizzani3
• Global-scale application by using 1
three satellite soil moisture products Research Institute for Geo-Hydrological Protection, National Research Council, Perugia, Italy, 2Department of Geodesy and
• The new satellite products estimate Geoinformation, Vienna University of Technology, Vienna, Austria, 3Institute of Atmospheric Sciences and Climate, National
accurately the accumulated rainfall Research Council, Bologna, Italy

Supporting Information:
• Readme
Abstract Measuring precipitation intensity is not straightforward; and over many areas, ground observations
• Figures S1–S8 are lacking and satellite observations are used to fill this gap. The most common way of retrieving rainfall is by
addressing the problem “top-down” by inverting the atmospheric signals reflected or radiated by atmospheric
Correspondence to: hydrometeors. However, most applications are interested in how much water reaches the ground, a problem
L. Brocca,
luca.brocca@irpi.cnr.it that is notoriously difficult to solve from a top-down perspective. In this study, a novel “bottom-up” approach is
proposed that, by doing “hydrology backward,” uses variations in soil moisture (SM) sensed by microwave
satellite sensors to infer preceding rainfall amounts. In other words, the soil is used as a natural rain gauge.
Citation:
Brocca, L., L. Ciabatta, C. Massari, Three different satellite SM data sets from the Advanced SCATterometer (ASCAT), the Advanced Microwave
T. Moramarco, S. Hahn, S. Hasenauer, Scanning Radiometer (AMSR-E), and the Microwave Imaging Radiometer with Aperture Synthesis are used to
R. Kidd, W. Dorigo, W. Wagner, and
obtain three new daily global rainfall products. The “First Guess Daily” product of the Global Precipitation
V. Levizzani (2014), Soil as a natural
rain gauge: Estimating global rainfall Climatology Centre (GPCC) is employed as main benchmark in the validation period 2010–2011 for determining
from satellite soil moisture data, the continuous and categorical performance of the SM-derived rainfall products by considering the 5 day
J. Geophys. Res. Atmos., 119,
5128–5141, doi:10.1002/
accumulated values. The real-time version of the Tropical Rainfall Measuring Mission (TRMM) Multisatellite
2014JD021489. Precipitation Analysis product, i.e., the TRMM-3B42RT, is adopted as a state-of-the-art satellite rainfall product.
The SM-derived rainfall products show good Pearson correlation values (R) with the GPCC data set, mainly in
Received 13 JAN 2014 areas where SM retrievals are found to be accurate. The global median R values (in the latitude band ±50°) are
Accepted 14 APR 2014
Accepted article online 15 APR 2014 equal to 0.54, 0.28, and 0.31 for ASCAT-, AMSR-E-, and SMOS-derived products, respectively. For comparison, the
Published online 6 MAY 2014 median R for the TRMM-3B42RT product is equal to 0.53. Interestingly, the SM-derived products are found
to outperform TRMM-3B42RT in terms of average global root-mean-square error statistics and in terms of
detection of rainfall events. The regions for which the SM-derived products perform very well are Australia,
Spain, South and North Africa, India, China, the Eastern part of South America, and the central part of the United
States. The SM-derived products are found to estimate accurately the rainfall accumulated over a 5 day period,
an aspect particularly important for their use for hydrological applications, and that address the difficulties of
estimating light rainfall from TRMM-3B42RT.

1. Introduction
Accurate estimates of rainfall are of vital importance for mitigation strategies of natural hazards such as floods
and landslides [Keefer et al., 1987; Hong et al., 2007; Wake, 2013; Hou et al., 2013] as well as for disease and
famine prevention [Scholthof, 2006] and many other applications. Nowadays, satellite rainfall products such as
the Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis product (TMPA) [Huffman
et al., 2007] are the main tools to address rainfall estimation on a global scale, particularly in data-poor world
areas where ground observations are sparse [Kidd and Levizzani, 2011]. The deployment of the Global
Precipitation Measurement mission satellite constellation, whose main spacecraft was launched on 27 February
2014, represents a critical advance in these efforts (http://pmm.nasa.gov/GPM) [Hou et al., 2013]. However, the
achievement of the accuracy levels required by the hydrological applications has proven difficult [Hossain and
Anagnostou, 2004; Uijlenhoet and Berne, 2008; Serrat-Capdevila et al., 2014]. Over land, this is mainly due to the
uncertainty in surface emissivity and its variability [Tapiador et al., 2012]. Currently, one of the major issues for
rainfall retrieval from space is related to the estimation of light rainfall that causes a general underestimation of
rainfall accumulations [Kucera et al., 2013]. These could be the main reasons for the continued prominence of in
situ gauge and precipitation radar observations in the majority of applications.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5128
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

An underexplored possibility for estimating rainfall lies in the more efficient use of complementary spaceborne
water cycle observations [McCabe et al., 2008]. Among them, soil moisture (SM) shares an obvious and strong
physical connection with rainfall. On the one hand, on a regional scale, antecedent SM has a significant impact
on the rainfall-triggering mechanism [Koster and the GLACE Team, 2004; Taylor et al., 2012], mainly for convective
storms. On the other hand, rainfall is the main driver of SM temporal variability locally [e.g., Crow et al., 2009].
After a rainfall pulse, SM exhibits a sudden increase followed by a smooth recession limb driven by
evapotranspiration and drainage [Porporato et al., 2004; Brocca et al., 2014]. Based on this concept, some recent
studies have shown that satellite SM data can be effectively employed for correcting satellite [Pellarin et al.,
2008; Crow et al., 2009; Pellarin et al., 2013] and reanalysis [Chen et al., 2012] rainfall products. These frameworks
need a first guess rainfall estimate (typically given by the real-time TMPA rainfall product) as the proposed
algorithms are only able to correct rainfall by using SM data. More recently, Brocca et al. [2013] developed an
approach (called SM2RAIN) based on the inversion of the hydrological water balance, for estimating, and not
correcting, rainfall from SM observations. The soil is regarded as a natural reservoir used for measuring the
amount of rainfall. The capability of SM2RAIN method to estimate, and not only correct, rainfall might be
particularly useful in those regions for which satellite rainfall data are affected by higher errors or not available as,
for instance, for the higher latitudes (>50°) in the case of TMPA products. Moreover, through SM2RAIN a fully
independent rainfall product is obtained characterized by a different error structure that will allow an optimal
integration with other satellite-based rainfall products through data assimilation techniques. The first application
of this approach to three sites in Italy (Umbria), Spain (Remedhus), and France (Valescure) with in situ and satellite
SM data has demonstrated its capability to estimate rainfall satisfactorily [Brocca et al., 2013]. Specifically, when
hourly in situ SM observations were used, the correlation coefficient, R, with daily rainfall observations ranged
between 0.900 and 0.945 for the three sites. When satellite data were considered, the R values against 5 day
rainfall data were found to be equal to 0.80 and 0.78 for the site in Italy and Spain, respectively. These results
encourage the application of the SM2RAIN method on a global scale. This is made possible by the availability of
several global satellite SM products developed in the last decades with good spatial-temporal resolution and
accuracy [e.g., Owe et al., 2008; Kerr et al., 2012a; Wagner et al., 2013]. Through the application of the SM2RAIN
algorithm, these products can be effectively used for rainfall estimation on a global scale.
On this basis, this study proposes a bottom-up rainfall measurement approach, SM2RAIN, as a new way of
monitoring rainfall from a satellite perspective. This approach is opposed to the classical top-down
perspective that is based on the inversion of the atmospheric signals reflected or radiated by atmospheric
hydrometeors. SM2RAIN is applied for terrestrial rainfall estimates on a global scale by using SM observations.
The capability of SM2RAIN to retrieve rainfall accumulations is expected to provide improvements in our
overall capability to estimate rainfall, complementing more classical approaches for estimating rainfall from
space. Three satellite SM products are used (1) the MetOp Advanced SCATterometer (ASCAT) [Wagner et al.,
2013], (2) the NASA Advanced Microwave Scanning Radiometer for EOS (AMSR-E) [Owe et al., 2008], and (3)
the Soil Moisture and Ocean Salinity (SMOS) Microwave Imaging Radiometer with Aperture Synthesis [Kerr
et al., 2012b]. From each SM data set, a new SM-derived rainfall product is obtained through the application
of the SM2RAIN algorithm. The quasi-global (±50° north-south latitude band) TRMM-3B42RT product is used
in the comparison as state-of-the-art rainfall product.
To evaluate the overall accuracy of the rainfall estimates, three rainfall products available on a global scale and at
daily time resolution are considered: the First Guess Daily product of the Global Precipitation Climatology Centre
(GPCC) which is based on ground observations only [Schamm et al., 2014], the One-Degree Daily precipitation
data set of the Global Precipitation Climatology Project (GPCP), based on ground plus satellite data [Huffman
et al., 2001], and the modeled 3 h rainfall data obtained from the ERA-Interim (ERAI) reanalysis [Dee et al., 2011].
The combined use of three products will allow a comprehensive and fair evaluation of the rainfall estimates
obtained from SM2RAIN. For ASCAT and AMSR-E (SMOS), the period 2007–2009 (2012) is used for the calibration
of SM2RAIN algorithm, the common period 2010–2011 is used for the validation of the rainfall products.

2. Soil Moisture and Rainfall Data Sets


2.1. Satellite Soil Moisture Observations
As mentioned above, three satellite SM products are used in this study. All the satellite SM products are
characterized by a nearly daily global coverage when both ascending and descending orbits are considered.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5129
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

The MetOp ASCAT SM data are obtained from C band (5.255 GHz, VV polarization) backscatter observations
using the TU-Wien algorithm water retrieval package at version 5.5 release 1.1 [Wagner et al., 2013]. The
algorithm is based on a time series change detection approach previously developed for the ERS-1/2
scatterometer by Wagner et al. [1999]. In this approach, SM is considered to have a linear relationship to
backscatter in the decibel space, the surface roughness is assumed to have a constant contribution in time,
and the typical yearly vegetation cycle is assumed to be known. The output of the algorithm is a time series
of relative SM percentage values for the first few centimeters of the soil (2–3 cm). The product is available
since 2007 with a spatial resolution of 25 km (resampled at 12.5 km).
The AMSR-E SM data set is obtained through the application of the VUA-NASA Land Parameter Retrieval Method
(LPRM) [Owe et al., 2008] at version 5 to C band (6.9 GHz) brightness temperature data. The LPRM algorithm is a
three-parameter retrieval model (SM, vegetation optical depth, and soil/canopy temperature) that uses the dual-
polarized channel (either 6.9 or 10.6 GHz) for the retrieval of surface SM and vegetation optical depth while the
land surface temperature is derived separately from the vertically polarized 36.5 GHz channel. The product is
expressed in volumetric terms (m3/m3) and is representative of a 2–3 cm deep surface soil layer. The product is
available from 2002 to October 2011 with a spatial resolution of 75 × 43 km (regridded to 0.25° regular grid).
For the SMOS data set [Kerr et al., 2012a], the reprocessed level 2 (L2) data generated with the latest available
version of the L2 algorithms (V5.51) are taken. The L Band Microwave Emission of the Biosphere Model
[Wigneron et al., 2007] is used for inverting SM from dual-polarized multiangular brightness temperature
observations of SMOS. The algorithm details can be found in Kerr et al. [2012b]. The product is expressed in
volumetric terms (m3/m3) and is representative of a soil layer of ~5 cm. The product is available since 2010
with a spatial resolution of 43 km.
For all the products, the quality flags supplied with the data are used to discard observations characterized by
frozen surface, temporary water and snow conditions, and high probability of radio frequency interference
(RFI). Moreover, retrievals characterized by a soil temperature (obtained from ERA-Interim reanalysis data)
below 3°C are masked out. After masking, the temporal resolution of satellite SM products is significantly
reduced in some regions and, hence, the 5 day accumulated rainfall data are considered in all the analyses.

2.2. Rainfall Data Sets


Four different rainfall data sets derived from satellite sensors, ground observations, and reanalysis are used.
For the sake of comparison, all the products are resampled at 1° spatial resolution and aggregated at 1 day
temporal resolution.
The baseline satellite rainfall product is the real-time version of the TMPA product [Huffman et al., 2007], i.e.,
the TRMM-3B42RT, available from 1997 onward with 3 h temporal resolution and a spatial resolution of 0.25°
for the ±50° north-south latitude band. Several studies [e.g., Ebert et al., 2007; Huffman et al., 2007] have
recognized that TRMM-3B42RT represents the state-of-the-art of the currently available satellite rainfall
products. Therefore, this product is used as comparison for assessing the possible benefits deriving from the
new SM-derived rainfall products.
The new First Guess Daily product provided by GPCC [Schamm et al., 2014], which is available since 1 January
2009 with a spatial sampling grid of 1°, was employed as main independent benchmark for the validation
period 2010–2011. The GPCP Daily Precipitation Data Set, version 1.2 [Huffman et al., 2001], which merges
data from over 6000 rain gauge stations with various types of satellite observations into a daily 1° resolution
rainfall product from October 1996 to the present, is used as benchmark in the calibration period 2007–2009
(2012) for ASCAT and AMSR-E (SMOS) (as the GPCC product is not available before 2009). Finally, the ERA-
Interim (ERAI) reanalysis rainfall data set [Dee et al., 2011], characterized by a spatial resolution of 0.703°, and
3 h temporal resolution is employed as an additional benchmark data set. For all rainfall products, rainfall and
snowfall variables are not distinguished in this study.

3. Methods
3.1. SM2RAIN Algorithm
The SM2RAIN algorithm is based on the inversion of the soil water balance equation for retrieving rainfall
from SM data. The soil is assumed to work as a natural rain gauge for measuring the amount of rainfall fallen

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5130
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Figure 1. Flowchart for the application of the SM2RAIN algorithm to satellite soil moisture data.

into the ground. Specifically, the soil water balance equation for a layer depth Z [L] can be described by the
following expression:
Zdsðt Þ=dt ¼ pðtÞ  r ðt Þ  eðtÞ  gðt Þ (1)
where s(t) [] is the relative saturation of the soil or relative SM, t [T] is the time and p(t), r(t), e(t),
and g(t) [L/T] are the precipitation, runoff, evapotranspiration, and drainage rate, respectively. Whenever
it rains, the evaporation rate can be safely assumed as negligible (e(t) = 0). Moreover, by assuming
that all precipitation infiltrates into the soil, the runoff rate is zero (r(t) = 0). For the drainage rate, the
following relation may be adopted, g(t) = as(t)b, where a [L/T] and b [] are two parameters expressing
the nonlinearity between drainage rate and soil saturation. The rearrangement of equation (1), with the
described assumptions, yields to
pðtÞ ≅ ZdsðtÞ=dt þ asðt Þb (2)

This equation can be used for estimating the precipitation rate from the knowledge of relative SM, s(t), its
fluctuations in time, ds(t)/dt, and three parameters (Z, a, and b) to be estimated through calibration. Even
though the assumptions made for deriving equation (2) might introduce some errors, they allow obtaining a
simple (but effective) method for the retrieval of rainfall from SM data [Brocca et al., 2013] that can be easily
applied on a global scale.
3.2. Rainfall Estimation Through SM2RAIN
The procedure used for estimating, through the SM2RAIN algorithm, daily rainfall at 1° resolution from each
SM data set is straightforward and is schematized as (see Figure 1)
1. Regrid SM data at 1° resolution by computing the average of the retrievals contained in each grid point.
The data are rescaled between 0 and 1 by using the maximum and minimum measurement values as
relative SM data should be used as input into the SM2RAIN algorithm and linearly interpolated at 0000
UTC of each day, for obtaining daily rainfall products equally distributed in time. A threshold of 4 days is
used for avoiding to interpolate the data when SM observations are not available.
2. Classify land points in several classes (12 for this study) in accordance with the amount of rainfall that is
estimated from the benchmark data set in the calibration period (GPCP in this study).
3. Calibrate the three SM2RAIN parameters (Z, a, and b) for each class by selecting a number of points
(equal to 400 for this study) not significantly affected by noise in the SM retrievals (i.e., excluding
the area covered by forest, desert, and characterized by complex topography). The GPCP product is
used here for the parameters calibration. The minimization of the root-mean-square error (RMSE),
averaged for each point of the rainfall class, is used as objective function; and the accumulated
rainfall over 5 days is considered.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5131
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

4. Run the SM2RAIN algorithm for all the land points with the optimized parameters for the whole period for
which SM data are available.
5. Mask rainfall retrievals when daily soil temperature is below a certain threshold and when the SM data set
is flagged (frozen surface, temporary water and snow conditions, and radio frequency interference).
We note here that several attempts are made for distributing the SM2RAIN parameters. Specifically, soil
(from the Harmonized World Soil Database v1.2), land use, and climatic regimes (Köppen-Geiger climate
classification) maps are selected but the best results are obtained by using rainfall classes. For the sake of
brevity, only this analysis is reported in the manuscript. Future investigations are planned to identify more in
detail the correlation between SM2RAIN parameters and other relevant properties at a finer spatial resolution
(e.g., soil, land use, topography, and climatic regimes).

3.3. Performance Metrics


Three continuous metrics are used for the evaluation of the performance of the SM-derived rainfall products
(and TRMM-3B42RT): the correlation coefficient, R, the root-mean-square error, RMSE, and the mean error,
BIAS, computed as percentage error between the accumulated rainfall of “observed” (GPCC, GPCP, and ERAI)
and estimated data in the investigated period.
Moreover, three categorical metrics, widely used for evaluating the performance of satellite rainfall data, are
also used: FAR (False Alarm Ratio), POD (Probability Of Detection), and TS (Threat Score). FAR refers to the
fraction of predicted events that are actually nonevents and POD to the fraction of all qualifying events
correctly predicted; TS provides an integrated measure of the overall performance. Through the definition of
a rainfall accumulation “event” as 5 day rainfall accumulation in excess of a given threshold, the categorical
metrics are defined as
F
FAR ¼ (3)
HþF
H
POD ¼ (4)
HþM
H
TS ¼ (5)
HþFþM
where H is the number of successfully predicted events by a given rainfall product, F the number of nonevents
erroneously predicted to occur, and M the number of actual events that are missed. In this evaluation, an event is
defined as a 5 day rainfall accumulation that exceeds a given percentile threshold of all 5 day accumulations
observed for a given 1° pixel in the analyzed period [Chen et al., 2012].

4. Results
Following the procedure schematized in Figure 1 and described above, the SM2RAIN parameters for each
satellite SM product are obtained. First, the three SM data sets are resampled on a regular 1° grid, thus
obtaining 13,237 points over land. Second, by using GPCP product as benchmark (only for the calibration
period), the land points are subdivided in 12 (nonequally spaced) classes as a function of the rainfall amount.
Specifically, the following thresholds, in terms of annual rainfall (in mm), are considered for deriving 12
classes: 0–133, 134–267, 268–367, 368–467, 468–567, 568–667, 668–767, 768–867, 868–1067, 1068–1333,
1333–1833, and >1834. The number of points of each class is nearly uniformly distributed and ranges
between 900 and 1400. Third, the SM data sets are splitted into a calibration (2007–2009 for ASCAT and
AMSR-E and 2012 for SMOS) and a validation (2010–2011, common to all products) period. Note that (results
not shown) calibrating ASCAT in 2012 provides nearly the same results. For each class, 400 points are selected
where the SM2RAIN parameters are calibrated (separately for each SM product). Fourth, the algorithm is run
for the whole period with the parameters spatially distributed according to the previously defined classes.
Finally, rainfall retrievals are masked by considering the quality flags of each SM product and a soil
temperature threshold of 3°C (from ERA-Interim reanalysis data). After masking, on average the percentage of
missing data is equal to 14%, 18%, and 21% for ASCAT, AMSR-E, and SMOS products, respectively.
For the sake of simplicity, in the sequel we will refer to the three SM-derived rainfall products by using only
the name of the satellite SM data set (i.e., ASCAT, AMSR-E, and SMOS) and to the TRMM-3B42RT product with

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5132
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

a
Table 1. SM2RAIN Parameter Values (Z, a, and b) Obtained From the Calibration Against the GPCP Product in the Calibration Period 2007–2009 for ASCAT
Rainfall Classes (mm/yr)

0–133 134–267 268–367 368–467 468–567 568–667 668–767 768–867 868–1067 1068–1333 1333–1833 >1834

Z (mm) 26.6 32.7 27.7 29.9 36.3 35.3 33.9 35.4 46.3 53.8 60.9 56.5
a (mm/d) 1.0 2.4 4.2 7.2 9.3 9.3 10.4 12.0 20.3 19.9 24.7 23.3
b () 23.2 1.5 1.8 3.0 3.4 2.8 2.4 2.4 2.6 2.3 2.5 1.4
a
Parameters are subdivided in 12 classes as a function of the annual rainfall amount obtained from GPCP.

TRMM-RT. Analogously, the three benchmark data sets are referred to with their acronyms: GPCC, GPCP,
and ERAI.
As an example, Table 1 shows the SM2RAIN parameter values (Z, a, and b) obtained from the calibration
against GPCP product in the calibration period 2007–2009 for ASCAT, subdivided for the 12 rainfall classes. As
it can be seen, the values of the parameters Z and a increase with rainfall as it can be expected from equation
(2) by considering that SM data are normalized between 0 and 1. The parameter b is found to be varying in a
lower range (1.4–3.4 if the first class is excluded). The anomalous a and b values obtained for the first class (0–
133 mm) are likely due to the noise of ASCAT SM retrievals, especially in desert areas, which produces
erroneous rainfall even where the observations are actually very low. The continuous performance in the
calibration period for the estimated rainfall from ASCAT is shown in Figure 2 and compared with those
obtained through the TRMM-RT product. We underline that some of the satellite sensors used for developing
the GPCP product are also used by the TRMM-RT algorithm. The ASCAT-derived SM product shows median R
values equal to 0.48 and 0.52 in the comparison against the GPCP and ERAI products, respectively. When
using ERAI, the correlation is slightly higher than that obtained with the TRMM-RT product (median R = 0.51).
In terms of RMSE, the ASCAT-derived product performs better than TRMM-RT for both benchmark data sets
while BIAS results are similar.
4.1. Continuous Performance Assessment
After the calibration, all the SM-derived products are evaluated in the validation period from 14 January 2010
to 30 September 2011 against the GPCC rainfall product. We underline here that the selection of a different
data set (GPCC rather than GPCP) and an independent time period maximizes the even-handedness of the
comparison against the TRMM-RT product. However, it is evident that if the TRMM-RT product were calibrated
on these data sets, better performance would be expected.
Figure 3 shows the corresponding global maps of correlation for the three rainfall products derived from the
three satellite SM data sets and for the TRMM-RT product. Note that the 5 day accumulated rainfall data
are considered in all the analyses made. The selection of the GPCC product (instead of GPCP) is motivated by
the need to perform a fair comparison when considering the TRMM-RT product (for the sake of completeness,
the comparison against the GPCP and ERAI products is shown in the supporting information, Figures S1
and S2). For each panel, the insets show the correlation values averaged as a function of the vegetation
optical depth (VOD), a measure of vegetation density obtained from the AMSR-E sensor [Owe et al., 2008].
Table 2 summarizes the main statistics of the scores used for the evaluation of the rainfall products.

Figure 2. Summary results in terms of correlation coefficient (R), root-mean-square error, RMSE, and BIAS for the estimated
rainfall from ASCAT, and the TRMM-RT product, against GPCP and ERAI products in the calibration period 2007–2009. The
results are computed for the ±50° latitude band (TRMM-RT coverage). In the box plots the min/max, 25/75° percentiles and
the median values are shown.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5133
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Figure 3. Correlation, R, maps of the 5 day rainfall products derived from ASCAT, AMSR-E, and SMOS soil moisture (through
SM2RAIN) and from TRMM-RT against the GPCC product used as benchmark in the validation period 2010–2011. The insets
show the histogram of R averaged as a function of the Vegetation Optical Depth (VOD). The median R values are also shown
(the value corresponding to the TRMM-RT data set, i.e., between ±50° latitude, are reported in brackets). The R values
greater than 0.20 (from yellow to red) are significant at the 0.01 confidence level.

Overall, the correlation between the estimated rainfall from SM data and the GPCC data set is remarkable and,
on average, comparable with the TRMM-RT product when ASCAT is considered (R = 0.54, see Figure 3), but
lower for AMSR-E- and SMOS-derived products. The lower performances of AMSR-E and SMOS are attributed
to their lower temporal resolution (mainly due to radio frequency interference, RFI) which significantly affects
the rainfall retrieval through SM2RAIN. Moreover, prior to July 2010, SMOS was in the commissioning phase
during which lower retrieval performance is expected [Kerr et al., 2012a]. Note that the spatial correlation
pattern shown in Figure 3 closely resembles the expected accuracy of the SM data sets [Dorigo et al., 2010;
Leroux et al., 2013]. Lower correlation values have to be expected over tropical forests where the signal of all
satellite sensors does not penetrate the dense vegetation cover (for SMOS these areas are masked out and no

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5134
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Table 2. Summary of Performance Scores (Median and Standard Deviation in Brackets) Computed for 5 Day Accumulated Rainfall From the Three Soil Moisture-
a
Derived Rainfall Products (ASCAT, AMSR-E, and SMOS) and for TRMM-RT Against the GPCC Product Used as Benchmark in the Validation Period 2010–2011
R RMSE (mm) BIAS ()
Rainfall
Product ALL VOD <0.5 VOD >0.5 ALL VOD <0.5 VOD >0.5 ALL VOD <0.5 VOD >0.5

ASCAT 0.47 0.48 0.42 11.76 10.38 19.62 0.14 0.14 0.18
(0.26) (0.27) (0.21) (9.28) (7.62) (9.69) (0.38) (0.38) (0.34)
AMSR-E 0.26 0.29 0.19 11.88 10.64 22.26 0.13 0.15 0.12
(0.23) (0.23) (0.20) (11.12) (8.96) (12.29) (0.43) (0.42) (0.44)
SMOS 0.24 0.28 0.09 11.57 10.80 8.94 0.13 0.14 0.15
(0.25) (0.25) (0.19) (10.13) (18.20) (12.02) (0.42) (0.42) (0.44)
TRMM-RT 0.53 0.53 0.55 16.36 14.47 25.38 0.04 0.06 0.07
(0.20) (0.20) (0.17) (11.78) (10.29) (12.43) (0.37) (0.38) (0.30)
a
The scores are shown for the whole domain (ALL) and also subdivided as a function of the vegetation optical depth, VOD. For the BIAS, the values outside the
range ±1 are masked out. (R: correlation coefficient, RMSE: root-mean-square error (mm).)

retrieval is done). A general decrease of the correlation with vegetation density (VOD) is observed for the
AMSR-E- and SMOS-derived products, while this is not the case for ASCAT. This finding is in accordance with
the expected behavior of passive (AMSR-E and SMOS) and active (ASCAT) microwave retrievals [Dorigo et al.,
2010]. Over desert areas, and particularly over the Sahara, the Namibian desert, and the Arabian Peninsula,
the ASCAT-derived product provides the lowest performance likely due to volume scattering effects from dry
subsurface soil layers [Wagner et al., 2013]. However, we note that in these regions also the gauge-based
rainfall products are affected by significant uncertainties. Similar issues are found for regions with very
complex topography as, for instance, the Andes and the Himalaya. At northern latitudes, the residual
presence of snow, ice, and freezing conditions negatively affects the correlation values for all the products.
Excluding these well-known challenging regions, covering less than 18% of the land points, a general very
good performance is reached elsewhere, particularly with ASCAT (median R = 0.578), and mainly over the
Australian continent, South Africa, Sahel, Brazil, Argentina, Mexico, central United States, Spain, India, and
China. In these areas, all the SM-derived products perform well with correlations frequently higher than
TRMM-RT (58% of cases for ASCAT). As to the comparison with the GPCP and ERAI products (Figures S1 and
S2), results of the SM-derived products are quite similar both in quantitative terms and for the spatial
correlation patterns thus corroborating their robustness (see also the supporting information). Moreover, a
consistent spatial pattern can be seen comparing the results shown in Figure 3 for AMSR-E with those
reported in Crow et al. [2010] (Figure 5) who used a similar, but different, approach for correcting rainfall
through SM data.
The accuracy of the rainfall products is also evaluated in terms of absolute amounts. The RMSE and the BIAS
are computed by using the GPCC (Figure 4 and Table 2) and the GPCP/ERAI products (Figure S3 in the
supporting information) as benchmark. The most interesting result comes out from Figure 4 for the RMSE.
The SM-derived rainfall products show very good performance with median RMSE equal to 11.8, 11.8, and
11.4 mm for ASCAT, AMSR-E, and SMOS, respectively, while TRMM-RT performs worst with median
RMSE = 16.4 mm. A similar picture is found when the GPCP/ERAI products are adopted (Figure S3). In order to
deeply investigate the motivations of these results, the temporal standard deviation of the 5 day rainfall time
series of each product is analyzed (see Figure S4 in the supporting information). A highly consistent behavior
between the three benchmark rainfall data sets (GPCC, GPCP and ERAI) is obtained with most land areas
characterized by values on the order of 5–15 mm. The SM-derived products are found to be less temporally
variable particularly for the SMOS and AMSR-E products while ASCAT shows a good agreement till to 80°
percentile and above it tends to underestimate. On the other hand, the TRMM-RT product has found to
overestimate the temporal variability and this might be an important reason for its highest RMSE values.
This result is clarified from the Taylor diagram [Taylor, 2001] shown in Figure S5 (supporting information)
where it is evident that the higher RMSE of TRMM-RT product is due to its higher temporal standard deviation
while the SMOS and AMSR-E products have lower RMSE due to the tendency of the SM-derived product to
overly suppress rainfall temporal variability.
In terms of BIAS, the rainfall products derived from AMSR-E and SMOS show a median value close to zero but
higher values (>30%) in absolute terms. The ASCAT-derived product shows a general overestimation but a

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5135
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Figure 4. Frequency histogram and box plots of the (left) root-mean-square error (RMSE) computed on 5 day accumulated
rainfall, and (right) BIAS computed on the accumulated rainfall (in %) for the three soil moisture-derived rainfall products
(ASCAT, AMSR-E, and SMOS) and for TRMM-RT against the GPCC data set used as benchmark in the validation period 2010–
2011. The RMSE (BIAS) median values and the standard deviation (median of absolute values) are shown for each product.
The values of the BIAS outside the range ±100% are masked out. In the box plots the min/max, 25/75° percentiles and the
median values are shown.

slightly lower median absolute value. The accumulated rainfall for each product in the validation period is
also computed and mapped in Figures S6 and S7 of the supporting information. It is obtained that all the SM-
derived products are able to reproduce very well the overall rainfall pattern (with some local differences)
even though the total amount might be quite different (see, for instance, at the higher latitudes for the AMSR-
E- and SMOS-derived products, Figure S7).

4.2. Categorical Performance Assessment


As mentioned above, the rainfall products are also evaluated in terms of categorical performance. Therefore,
FAR, POD, and TS are computed for the different SM-derived rainfall products, and for TRMM-RT (see Figure 5
for GPCC and Figure S8 for GPCP and ERAI in the supporting information). In terms of FAR, TRMM-RT
outperforms the other three products likely due to the positive noise in the SM time series that are
interpreted by SM2RAIN as a low-intensity rainfall event (as already highlighted in Crow et al. [2011]).
Conversely, in terms of POD, the SM-derived products tend to outperform TRMM-RT. This behavior is
attributed to the difficulties of estimating light rainfall from TRMM-RT that appears to be addressed by the
proposed approach. By considering the TS, ASCAT outperforms the other products for lower rainfall
percentiles (<40 percentiles) while TRMM-RT is slightly better for the higher percentiles thus confirming the
tendency of the SM2RAIN algorithm to underestimate the higher rainfall rates [Brocca et al., 2013]. The
comparison against the GPCP and ERAI products provides very similar results (Figure S8).

Figure 5. Global averages of categorical metrics, computed on 5 day accumulated rainfall, for the three soil moisture-
derived rainfall products (ASCAT, AMSR-E, and SMOS) and for TRMM-RT against the GPCC data set used as benchmark in
the validation period 2010–2011, (left) False Alarm Ratio, (middle) Probability of Detection, and (right) Threat Score for a
range of 5 day rainfall accumulation threshold. An event is defined as a 5 day rainfall accumulation that exceeds a given
percentile threshold of all 5 day accumulations observed for a given 1° pixel over the whole analyzed period. For each
percentile, the mean values ±1 standard deviation are shown as solid and dashed lines, respectively.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5136
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Figure 6. As in Figure 3 but for the Threat Score (TS) computed considering a rainfall threshold of 1 mm.

Figure 6 shows the global map of TS for the different products against the GPCC product in the validation
period. For this figure, a rainfall event is defined as a 5 day rainfall accumulation in excess of 1 mm. The maps
clearly highlight the good performance of the SM-derived products to detect rainfall events. Considering the
±50° latitude region, the median TS is found to be equal to 0.808, 0.806, and 0.606 for ASCAT, AMSR-E, and
SMOS, respectively. The corresponding value for TRMM-RT is equal to 0.660.

4.3. Products Assessment at Monthly Time Scale


The SM-derived products are also evaluated at the monthly time scale by using the high-quality GPCC Full
Data Reanalysis of monthly global land surface precipitation based on the 67,200 stations worldwide
[Schneider et al., 2014]. As this product is available until December 2010, the comparison was made for the
period 2007–2010 even though this means that for the SMOS-derived product only the year 2010 is

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5137
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Figure 7. Summary results in terms of correlation coefficient (R), root-mean-square error, RMSE, and BIAS for the estimated
rainfall from ASCAT, AMSR-E, and SMOS and the TRMM-RT product, against the high-quality monthly product from GPCC in
the period 2007–2010. The results are computed for the ±50° latitude band (TRMM-RT coverage). For SMOS, the results
refer to the year 2010 only. In the box plots the min/max, 25/75° percentiles and the median values are shown.

considered. Figure 7 summarizes the results in terms of box plots for R, RMSE, and BIAS. As it can be seen, the
previous results are confirmed also by using the high-quality GPCC monthly product. Specifically, for ASCAT
the median R value is equal to 0.68 and close to those obtained with TRMM-RT (R = 0.70) while the median
RMSE = 39.88 mm for ASCAT is significantly lower than those obtained with the other products
(RMSE > 45 mm).
In order to visualize the rainfall time series, the monthly rainfall values from January 2007 to December 2009
are reported in Figure 8 for the ASCAT and TRMM-RT products. Specifically, 100 points (out of 13237 land
points) evenly distributed on a quasi-global scale are displayed and ordered by latitude and longitude for

Figure 8. Monthly rainfall time series obtained from the ASCAT SM-derived rainfall products and from the TRMM-RT against the GPCC data set for 100 points nearly
evenly distributed on a quasi-global scale (Lon = 90:150, Lat = 50:50) from January 2007 to December 2009. Note that the y axis is truncated at 350 mm for the
sake of visualization.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5138
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Figure 9. Rainfall time series for 5 day accumulated rainfall obtained from the ASCAT SM-derived products and from the TRMM-RT against the ERAI data set for six
points in Brazil, South Africa, Eastern Australia, United States, Spain, and China from 2007 to 2011.

giving a visual interpretation of the analysis. As expected, the ASCAT-derived product slightly underestimates
rainfall in the Tropical and Equatorial bands (see middle rows). However, elsewhere a general agreement
between the seasonal patterns for each point and the total amount is observed with a general better
agreement for the high latitudes (first two rows).

4.4. Time Series for Some Specific Locations


Finally, to visualize the data at 5 day temporal resolution, a time series plot for some specific locations
worldwide (in Brazil, South Africa, Eastern Australia, United States, Spain and China) is shown in Figure 9 for
the ASCAT and TRMM-RT products against the ERAI data set and by considering the 5 year period 2007–2011.
It is evident the reliability of the ASCAT-derived rainfall product in reproducing the seasonal and long-term
rainfall pattern with results comparable with the TRMM-RT product.

5. Conclusions
The results shown in this study offer compelling evidence that it is possible to estimate rainfall from satellite
SM data on a global scale (also for high latitudes). Specifically, by considering the daily GPCC product as
benchmark in the validation period 2010–2011 the following conclusions can be drawn:

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5139
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

1. the global correlation maps (Figure 3) shows that the SM-derived rainfall products provide good R values
in the areas where SM retrievals are expected to be accurate. In the latitude band ±50°, the median
R values are equal to 0.542, 0.281, and 0.313 for the ASCAT, AMSR-E, and SMOS products, respectively.
The corresponding R value for the TRMM-RT product is equal to 0.534. The regions for which the SM-
derived products perform very well are Australia, Spain, South and North Africa, India, China, the Eastern
part of South America, and the central part of the United States.
2. the most interesting results are obtained in terms of RMSE (Figure 4). In fact, the median RMSE values are
lower than 12 mm while for the TRMM-RT product a median value equal to 16.4 mm is obtained. In terms
of BIAS, the results of the three SM-derived products are quite similar and slightly worse than the TRMM-
RT product.
3. the categorical scores (Figures 5 and 6), and specifically the TS, show that the SM-derived product
performs well in the detection of low to medium rainfall events.
4. the analysis at the monthly time scale (Figures 7 and 8) with the higher quality GPCC product confirms the
previous results in terms of continuous performance scores.
Overall, these results demonstrate the validity of the proposed SM2RAIN method for improving the rainfall
estimation on a global scale with a promising reduction of the RMSEs and a better detection of low to
medium rainfall events with respect to the TRMM-RT product. However, we acknowledge that the reduction
into the RMSE values of the SM-derived products might be linked to their tendency to overly suppress rainfall
accumulation variability. Further improvements can be obtained by merging different satellite SM products,
both in space and in time, and by refining the parameterization of the SM2RAIN. Additionally, new satellite
missions are foreseen for the remote sensing of SM with expected improved accuracy and spatial-temporal
resolution such as the Soil Moisture Active and Passive mission that will be launched in October 2014
[Entekhabi et al., 2010; Brown et al., 2013]. The proposed approach has thus still room for improvement and it
will become an essential new tool for global-scale rainfall estimation over land surfaces. Moreover, the same
approach can be used to evaluate the reliability of different satellite SM products on a regional and global
scale by using rainfall data as benchmark [Crow et al., 2010; Tuttle and Salvucci, 2014]. This would allow
addressing the issue of the validation of coarse scale SM products due to the lack of dense in situ SM network
[Wagner et al., 2014].

Acknowledgments References
The authors gratefully acknowledge
Brocca, L., F. Melone, T. Moramarco, and W. Wagner (2013), A new method for rainfall estimation through soil moisture observations,
support from the ESA Climate Change
Geophys. Res. Lett., 40, 853–858, doi:10.1002/grl.50173.
Initiative (CCI) (ESA/ESRIN contract
Brocca, L., S. Camici, F. Melone, T. Moramarco, J. Martinez-Fernandez, J.-F. Didon-Lescot, and R. Morbidelli (2014), Improving the represen-
4000104814/11/I-NB, http://www.esa-
tation of soil moisture by using a semi-analytical infiltration model, Hydrol. Processes, 28(4), 2103–2115.
soilmoisture-cci.org/), EUMETSAT
Brown, M. E., V. Escobar, S. Moran, D. Entekhabi, P. E. O’Neill, E. G. Njoku, B. Doorn, and J. K. Entin (2013), NASA’s Soil Moisture Active Passive
through the “Satellite Application
(SMAP) mission and opportunities for applications users, Bull. Am. Meteorol. Soc., 94, 1125–1128.
Facility on Support to Operational
Chen, F., W. T. Crow, and T. H. Holmes (2012), Improving long-term, retrospective precipitation datasets using satellite-based surface soil
Hydrology and Water Management
moisture retrievals and the soil moisture analysis rainfall tool, J. Appl. Remote Sens., 6(1), 063604.
(H-SAF)” (http://hsaf.meteoam.it/), and
Crow, W. T., G. F. Huffman, R. Bindlish, and T. J. Jackson (2009), Improving satellite rainfall accumulation estimates using spaceborne soil
the European Commission through
moisture retrievals, J. Hydrometeorol., 10, 199–212.
the FP7 Project eartH2Observe. S.
Crow, W. T., D. G. Miralles, and M. H. Cosh (2010), A quasi-global evaluation system for satellite-based surface soil moisture retrievals, IEEE
Camici, A. Mondini, M. Rossi, and A.
Trans. Geosci. Remote Sens., 48(6), 2516–2527.
Tarpanelli are thanked for their useful
Crow, W. T., M. J. van Den Berg, G. F. Huffman, and T. Pellarin (2011), Correcting rainfall using satellite-based surface soil moisture retrievals:
comments on an earlier version of
The Soil Moisture Analysis Rainfall Tool (SMART), Water Resour. Res., 47, W08521, doi:10.1029/2011WR010576.
this paper. We also thank Wade Crow
Dee, D. P., et al. (2011), The ERA-Interim reanalysis: Configuration and performance of the data assimilation system, Q. J. R. Meteorolog. Soc.,
and two anonymous reviewers for
137, 553–597.
their useful suggestions that signifi-
Dorigo, W. A., K. Scipal, R. M. Parinussa, Y. Y. Liu, W. Wagner, R. A. M. de Jeu, and V. Naeimi (2010), Error characterisation of global active and
cantly improved the readability of the
passive microwave soil moisture datasets, Hydrol. Earth Syst. Sci., 14, 2605–2616.
paper. The rainfall data sets obtained
Ebert, E. E., J. Janowiak, and C. Kidd (2007), Comparison of near real time precipitation estimates from satellite observations and numerical
from ASCAT, AMSR-E, and SMOS at 1°
models, Bull. Am. Meteorol. Soc., 88, 47–64.
resolution are available at http://
Entekhabi, D., et al. (2010), The Soil Moisture Active and Passive (SMAP) mission, Proc. IEEE, 98(5), 704–716.
hydrology.irpi.cnr.it/people/l.brocca
Hong, Y., R. F. Adler, and G. J. Huffman (2007), An experimental global prediction system for rainfall triggered landslides using satellite
or by contacting the first author at
remote sensing and geospatial datasets, IEEE Trans. Geosci. Remote Sens., 45, 1671–1680.
luca.brocca@irpi.cnr.it.
Hossain, F., and E. N. Anagnostou (2004), Assessment of current passive-microwave and infrared-based satellite rainfall remote sensing for
flood prediction, J. Geophys. Res., 109, D07102, doi:10.1029/2003JD003986.
Hou, A. Y., R. K. Kakar, S. Neeck, A. A. Azarbarzin, C. D. Kummerow, M. Kojima, R. Oki, K. Nakamura, and T. Iguchi (2013), The Global
Precipitation Measurement (GPM) mission, Bull. Am. Meteorol. Soc., doi:10.1175/BAMS-D-13-00164.1, in press.
Huffman, G. J., R. F. Adler, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind (2001), Global precipitation at one-degree
daily resolution from multi-satellite observations, J. Hydrometeorol., 2, 36–50.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5140
Journal of Geophysical Research: Atmospheres 10.1002/2014JD021489

Huffman, G. J., R. F. Adler, D. T. Bolvin, G. Gu, E. J. Nelkin, K. P. Bowman, Y. Hong, E. F. Stocker, and D. B. Wolff (2007), The TRMM Multisatellite
Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales, J. Hydrometeorol., 8(1),
38–55.
Keefer, D. K., R. C. Wilson, R. K. Mark, E. E. Brabb, W. M. Brown, S. D. Ellen, E. L. Harp, G. F. Wieczoreck, C. S. Alger, and R. S. Zatkin (1987), Real-
time landslide warning during heavy rainfall, Science, 238, 921–925.
Kerr, Y. H., et al. (2012a), The SMOS soil moisture retrieval algorithm, IEEE Trans. Geosci. Remote Sens., 50(5), 1384–1403.
Kerr, Y. H., J. Font, M. Martin-Neira, and S. Mecklenburg (2012b), Introduction to the special issue on the ESA’s Soil Moisture and Ocean Salinity
Mission (SMOS) - Instrument performance and first results, IEEE Trans. Geosci. Remote Sens., 50(5), 1351–1353.
Kidd, C., and V. Levizzani (2011), Status of satellite precipitation retrievals, Hydrol. Earth Syst. Sci., 15, 1109–1116.
Koster, R. D., and the GLACE Team (2004), Regions of strong coupling between soil moisture and precipitation, Science, 305, 1138–1140.
Kucera, P. A., E. E. Ebert, F. J. Turk, V. Levizzani, D. Kirschbaum, F. J. Tapiador, A. Loew, and M. Borsche (2013), Precipitation from space:
Advancing earth system science, Bull. Am. Meteorol. Soc., 94, 365–375.
Leroux, D. J., Y. H. Kerr, P. Richaume, and R. Fieuzal (2013), Spatial distribution and possible sources of SMOS errors at the global scale, Remote
Sens. Environ., 133, 240–250.
McCabe, M., E. F. Wood, R. Wojcik, M. Pan, J. Sheffield, H. Gao, and H. Su (2008), Hydrological consistency using multi-sensor remote sensing
data for water and energy cycle studies, Remote Sens. Environ., 112, 430–444.
Owe, M., R. A. M. De Jeu, and T. R. H. Holmes (2008), Multi-sensor historical climatology of satellite-derived global land surface moisture,
J. Geophys. Res., 113, F01002, doi:10.1029/2007JF000769.
Pellarin, T., A. Ali, F. Chopin, I. Jobard, and J.-C. Bergs (2008), Using spaceborne surface soil moisture to constrain satellite precipitation
estimates over West Africa, Geophys. Res. Lett., 35, L02813, doi:10.1029/2007GL032243.
Pellarin, T., S. Louvet, C. Gruhier, G. Quantin, and C. Legout (2013), A simple and effective method for correcting soil moisture and precipi-
tation estimates using AMSR-E measurements, Remote Sens. Environ., 136, 28–36.
Porporato, A., E. Daly, and I. Rodriguez-Iturbe (2004), Soil water balance and ecosystem response to climate change, Amer. Nat., 164(5),
625–632.
Schamm, K., M. Ziese, A. Becker, P. Finger, A. Meyer-Christoffer, U. Schneider, M. Schröder, and P. Stender (2014), Global gridded precipitation
over land: A description of the new GPCC First Guess Daily product, Earth Syst. Sci. Data, 6, 49–60.
Schneider, U., A. Becker, P. Finger, A. Meyer-Christoffer, M. Ziese, and B. Rudolf (2014), GPCC’s new land surface precipitation climatology
based on quality-controlled in situ data and its role in quantifying the global water cycle, Theor. Appl. Climatol., 115(1–2), 15–40.
Scholthof, K. B. G. (2006), The disease triangle: Pathogens, the environment and society, Nat. Rev. Microbiol., 5(2), 152–156.
Serrat-Capdevila, A., J. B. Valdes, and Z. S. Eugene (2014), Water management applications for satellite precipitation products: Synthesis and
recommendations, J. Am. Water Resour. Assoc., 50(2), 509–525.
Tapiador, F. J., et al. (2012), Global precipitation measurement: Methods, datasets and applications, Atmos. Res., 104–105, 70–97.
Taylor, C. M., R. A. de Jeu, F. Guichard, P. P. Harris, and W. A. Dorigo (2012), Afternoon rain more likely over drier soils, Nature, 489(7416),
423–426.
Taylor, K. E. (2001), Summarizing multiple aspects of model performance in a single diagram, J. Geophys. Res., 106(D7), 7183–7192,
doi:10.1029/2000JD900719.
Tuttle, S. E., and G. D. Salvucci (2014), A new approach for validating satellite estimates of soil moisture using large-scale precipitation:
Comparing AMSR-E products, Remote Sens. Environ., 142, 207–222.
Uijlenhoet, R., and A. Berne (2008), Stochastic simulation experiment to assess radar rainfall retrieval uncertainties associated with
attenuation and its correction, Hydrol. Earth Syst. Sci., 12, 587–601.
Wagner, W., G. Lemoine, and H. Rott (1999), A method for estimating soil moisture from ERS scatterometer and soil data, Remote Sens.
Environ., 70, 191–207.
Wagner, W., et al. (2013), The ASCAT soil moisture product: Specifications, validation results, and emerging applications, Meteorol. Z., 22(1), 5–33.
Wagner, W., et al. (2014), Clarifications on the “Comparison between SMOS, VUA, ASCAT, and ECMWF soil moisture products over four
watersheds in U.S.”, IEEE Trans. Geosci. Remote Sens., 52(3), 1901–1906.
Wake, B. (2013), Flooding costs, Nat. Clim. Change, 3, 778.
Wigneron, J. P., et al. (2007), L-band Microwave Emission of the Biosphere (L-MEB) Model: Description and calibration against experimental
data sets over crop fields, Remote Sens. Environ., 107(4), 639–655.

BROCCA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 5141

You might also like