Long-term gridded precipitation products are crucial for several applications in hydrology, agriculture and climate sciences. Currently available precipitation products suffer from space and time inconsistency due to the non-uniform density of ground networks and the difficulties in merging multiple satellite sensors. The recent “bottom-up” approach that exploits satellite soil moisture observations for estimating rainfall through the SM2RAIN (Soil Moisture to Rain) algorithm is suited to build a consistent rainfall data record as a single polar orbiting satellite sensor is used.
Here we exploit the Advanced SCATterometer (ASCAT) on board three Meteorological Operational (MetOp) satellites, launched in 2006, 2012, and 2018, as part of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Polar System programme. The continuity of the scatterometer sensor is ensured until the mid-2040s through the MetOp Second Generation Programme. Therefore, by applying the SM2RAIN algorithm to ASCAT soil moisture observations, a long-term rainfall data record will be obtained, starting in 2007 and lasting until the mid-2040s. The paper describes the recent improvements in data pre-processing, SM2RAIN algorithm formulation, and data post-processing for obtaining the SM2RAIN–ASCAT quasi-global (only over land) daily rainfall data record at a 12.5 km spatial sampling from 2007 to 2018. The quality of the SM2RAIN–ASCAT data record is assessed on a regional scale through comparison with high-quality ground networks in Europe, the United States, India, and Australia. Moreover, an assessment on a global scale is provided by using the triple-collocation (TC) technique allowing us also to compare these data with the latest, fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA5), the Early Run version of the Integrated Multi-Satellite Retrievals for Global Precipitation Measurement (IMERG), and the gauge-based Global Precipitation Climatology Centre (GPCC) products.
Results show that the SM2RAIN–ASCAT rainfall data record performs relatively well at both a regional and global scale, mainly in terms of root mean square error (RMSE) when compared to other products. Specifically, the SM2RAIN–ASCAT data record provides performance better than IMERG and GPCC in data-scarce regions of the world, such as Africa and South America. In these areas, we expect larger benefits in using SM2RAIN–ASCAT for hydrological and agricultural applications. The limitations of the SM2RAIN–ASCAT data record consist of the underestimation of peak rainfall events and the presence of spurious rainfall events due to high-frequency soil moisture fluctuations that might be corrected in the future with more advanced bias correction techniques.
The SM2RAIN–ASCAT data record is freely available at
Rainfall is ranked the first among the Essential Climate Variables by the Global Climate Observing System (GCOS) as it represents the most important variable in many applications in the geosciences (Maggioni and Massari, 2018). Long-term rainfall records are essential for drought monitoring (e.g. Forootan et al., 2019), water resource management (e.g. Abera et al., 2017), and climate studies (e.g. Herold et al., 2016; Pendergrass and Knutti, 2018), while near-real-time rainfall data are needed for the mitigation of the impacts of natural disasters such as floods and landslides (e.g. Wang et al., 2017; Camici et al., 2018; Brunetti et al., 2018; Kirschbaum and Stanley, 2018). Additional applications in which near-real-time rainfall plays a crucial role are weather forecasting, agricultural planning, and vector-borne and waterborne disease monitoring (e.g. Rinaldo et al., 2012; Thaler et al., 2018).
Three different techniques can be used for estimating rainfall: ground measurements, meteorological modelling, and remote sensing. Ground measurements are based on rain gauges and meteorological radars (Lanza and Vuerich, 2009), but also new approaches such as microwave links are being developed (e.g. Overeem et al., 2011). These measurements guarantee high accuracy, but they suffer in many regions from limited spatial coverage (Kidd et al., 2017). Alternatively, meteorological models are used to estimate rainfall mainly in areas without reliable ground observations (Ebert et al., 2007), e.g. reanalysis. The uncertainties associated with these estimates can be large, mainly in areas where ground observations are scarce (Massari et al., 2017a). Therefore, to fill the gaps in the spatial coverage of ground measurements and to improve the estimates obtained by models, different remote-sensing techniques have been developed in the last 30 years (Hou et al., 2014). The standard methods for estimating rainfall from space are based on instantaneous measurements obtained from microwave radiometers, radars, and infrared sensors (Kidd and Levizzani, 2011). These methods are based on inversion techniques where the upwelling radiation (or backscattered signal for radars) is related to the surface precipitation rate, i.e. a “top-down” approach (Brocca et al., 2014).
The most recent and successful example of satellite precipitation estimates
is represented by the Integrated Multi-Satellite Retrievals for Global
Precipitation Measurement (GPM) (IMERG) of the GPM mission (Hou
et al., 2014), which provide a high spatial (0.1
In recent years, a new “bottom-up” approach has emerged that uses satellite soil moisture observations to infer, or to correct, rainfall over land (Brocca et al., 2013a; Crow et al., 2009; Pellarin et al., 2013; Wanders et al., 2015). The major difference between the bottom-up and top-down approaches is in the type of measurement, i.e. accumulated rainfall rates with the bottom-up method and instantaneous rainfall rates with the top-down method. This difference makes the two approaches highly complementary and their integration has already been successfully tested and demonstrated in several recent studies (e.g. Brocca et al., 2016; Ciabatta et al., 2017; Chiaravallotti et al., 2018; Massari et al., 2019). When accumulated rainfall estimates are needed (e.g. daily rainfall), the bottom-up approach has the advantage of requiring a much lower number of measurements and, hence, of satellite sensors. The limitations of the bottom-up approach are the possibility to estimate only terrestrial rainfall and its dependence on land characteristics (e.g. low accuracy for dense vegetation coverage and complex topography; Brocca et al., 2014).
The bottom-up approach has been applied over a range of scales: global (Crow et al., 2011; Brocca et al., 2014; Ciabatta et al., 2018), continental (Wanders et al., 2015; Brocca et al., 2016), and local (Massari et al., 2014; Brocca et al., 2015; Román-Cascón et al., 2017). Moreover, different satellite soil moisture products have been considered including the SMOS (Soil Moisture Ocean Salinity mission; Brocca et al., 2016), ASCAT (Advanced SCATterometer; Brocca et al., 2017), AMSR-E (Advanced Microwave Scanning Radiometer; Crow et al., 2009), and SMAP (Soil Moisture Active and Passive; Koster et al., 2016; Tarpanelli et al., 2017; Zhang et al., 2019). The first studies employing satellite rainfall estimates obtained through the bottom-up approach for hydrological and water resources applications have been recently published (e.g. Ciabatta et al., 2016; Abera et al., 2017; Brunetti et al., 2018; Camici et al., 2018). These studies have highlighted the large potential of this technique as a complementary and useful approach for estimating rainfall from space, and they have also shown its main limitations. Specifically, the temporal resolution and the accuracy of satellite soil moisture products play a fundamental role in determining the accuracy of the bottom-up rainfall estimates.
In this study, we describe the newly developed SM2RAIN–ASCAT (Soil Moisture to Rain) rainfall data
record covering the period 2007–2018 and characterized by a spatial and temporal
sampling of 12.5 km d
The purpose of this study is twofold. As a first objective, we have applied
SM2RAIN algorithm at 1009 uniformly distributed points (with a spacing of
1.5
We underline that the paper goal is to present and describe the SM2RAIN–ASCAT quasi-global rainfall data record and to perform a comparison with state-of-the-art global rainfall products. We do not want to show a comprehensive assessment of the product. Indeed, we believe that researchers other than the product developers should perform the validation of the dataset. Even better, we stress the importance of performing the validation by using the datasets in hydrological or agricultural applications (e.g. flood prediction and agricultural water management).
Nine different datasets have been collected for this study, which are based on remote sensing, ground observations, and reanalysis. Refer to Table 1 for a summary of the datasets.
List of satellite, ground-based, and reanalysis products used in this study (the spatial and temporal sampling used in this study is reported).
The main input dataset for producing the SM2RAIN–ASCAT data record is the ASCAT
soil moisture data record provided by the EUMETSAT H SAF (
Three datasets obtained from the latest reanalysis of the ECMWF, i.e. ERA5,
have been used. The ERA5 reanalysis is characterized by a spatial resolution of
Ground-based rainfall datasets from regional networks have been also
collected including the Climate Prediction Center (CPC) Unified Gauge-Based
Analysis of Daily Precipitation in the United States, the gridded rainfall
data provided by
The ERA5 and local rainfall datasets have been regridded over the ASCAT grid (12.5 km) through the nearest neighbouring method and resampled at a daily timescale as accumulated rainfall from 00:00 to 23:59 UTC. The ERA5 evaporation and soil temperature data are also regridded to the same grid and aggregated at a daily scale as accumulated and average values from 00:00 to 23:59 UTC, respectively.
For the global assessment of SM2RAIN–ASCAT, two additional rainfall datasets
have been considered: the Global Precipitation Climatology Centre Full
Data Daily Product (Schamm et al., 2015) and GPM IMERG Early
Run product (Hou et al., 2014), hereinafter referred to as
GPM-ER. Due to the availability of GPM-ER from April 2014, the global
analysis has been carried out in the 4-year period from January 2014 to
December 2018. Moreover, for the global inter-comparison all the datasets
(SM2RAIN–ASCAT, ERA5, GPCC, and IMERG-ER) have been regridded at a
0.25
In the following, the methodology used for obtaining the SM2RAIN–ASCAT data record is described. Specifically, three steps are carried out (see Fig. 1): (1) surface soil moisture data pre-processing, (2) employing the SM2RAIN algorithm, and (3) rainfall data post-processing. Different configurations for the data pre- and post-processing and for the SM2RAIN model equation are considered; the details are given in Table 2.
Processing steps for obtaining the SM2RAIN–ASCAT global rainfall data record (2007–2018) from ASCAT surface soil moisture data: pre-processing, SM2RAIN algorithm, and post-processing. Each bullet represents a possible configuration that has been tested; the selected configuration is in a red, bold font.
Configurations used in the paper (SWI – soil water index,
BCO – Brooks–Corey, VGEN – van Genuchten, MUA – Mualem–van Genuchten, SWI-Tvar –
SWI with
The ASCAT surface soil moisture product is provided as relative soil moisture (between 0 and 1) at the overpass time of the satellite sensor (see Fig. A1 in the Appendix) for the mean daily revisit time of ASCAT. The period 2007–2012 only had MetOp-A data, while the period 2013–2018 had data from both MetOp-A and MetOp-B. For the application of the SM2RAIN algorithm, data should be equally spaced in time, and hence, we have linearly interpolated in time soil moisture observations every 24, 12 and 8 h. The interpolation may increase the risk of false rainfall events, but it is a required step to obtain accumulated rainfall over a fixed duration. In a preliminary test (not shown for brevity), we tested the three sampling frequencies with the baseline formulation for SM2RAIN (Eq. 6, see below). The best performance results were obtained with a 12 h sampling, particularly from 2013 to 2018, in which both MetOp-A and MetOp-B are available. Therefore, 12 h sampling has been used in the following analyses. The 24 h accumulated rainfall is obtained by summing the two 12 h accumulated rainfall datasets obtained for each day.
One of the major problems in using satellite soil moisture observations for
rainfall estimation is related to the high-frequency fluctuations caused by
measurement and retrieval errors. If positive, such fluctuations are
interpreted erroneously as rainfall by the SM2RAIN algorithm. Therefore,
satellite surface soil moisture data need to be filtered before being used
as an input into SM2RAIN. In previous studies, the exponential filtering has
been considered (Wagner et al., 1999). The exponential filter,
also known as the soil water index (SWI), has been used for filtering surface
soil moisture time series as a function of a single parameter,
For all the filtering approaches, the parameter values of the filters have been optimized point-by-point in order to reproduce the reference rainfall observations.
The SM2RAIN algorithm is based on the inversion of the soil water balance
equation and allows to estimate the amount of water entering the soil by
using soil moisture observations from in situ or satellite sensors as an input
(e.g. Brocca et al., 2013a, 2014, 2015; Koster et
al., 2016; Ciabatta et al., 2017; Massari et al.,
2014). Specifically, the soil water balance equation can be described by
the following equation (over non-irrigated areas):
For estimating the rainfall rate, Eq. (1) is applied
only during rainfall periods, and, hence, some of the components of the
equation can be considered as negligible. For instance, the actual
evapotranspiration rate during rainfall is quite low due to the presence of
clouds and, hence, the absence of solar radiation. Similarly, the surface
runoff rate, i.e. the water that does not infiltrate into the soil and
flows at the surface to the watercourses, is much lower than the rainfall
rate, mainly if Eq. (1) is applied at a coarse spatial
resolution (20 km), i.e. with satellite soil moisture data. Indeed, most of
the water becomes runoff flowing in the subsurface, and also the part that does
not infiltrate, due to for instance impervious land cover or soil, may
re-infiltrate downstream within a pixel at a 20 km scale. We have indirectly
tested this hypothesis by counting the number of days the ASCAT soil
moisture product is higher than the 99.5 percentile for 2 (or more)
consecutive days in the period 2007–2018. We have indirectly tested this
hypothesis by counting the number of days the ASCAT soil moisture product is
higher than the 99.5 percentile for 2 (or more) consecutive days in the period
2007–2018. We have found that the number of consecutive days in which the
soil is saturated is equal to 4 d (median value on a global scale) over
12 years, with 90 % of the land pixels with values lower than 12 d (i.e. 1 d yr
Following the indications obtained in Brocca et al. (2015), we
have assumed that the surface runoff rate, sr(
The actual evapotranspiration rate has been considered as an additional
input, together with soil moisture, here obtained from the ECMWF reanalysis,
ERA5.
Moreover, we have considered an additional formulation in which
Accordingly, we have used different formulations for Eq. (2) that are
compared with the baseline equation used in previous studies (e.g.
Brocca et al., 2014).
SM2RAIN parameter values are calibrated point-by-point by using the reference rainfall as the target. As an objective function, we have used the minimization of the root mean square error (RMSE) between the SM2RAIN–ASCAT and reference rainfall.
The use of satellite soil moisture observations for obtaining rainfall estimates is affected by errors in the input data and in the retrieval algorithm SM2RAIN. The correction of the overall bias in the climatology is a simple and effective approach for mitigating a part of such errors. Specifically, we refer here to a static correction procedure, which once calibrated for a time period can be applied in future periods and for operational real-time productions. We note that a climatological correction is performed in several satellite rainfall datasets delivered in near real time (e.g. GPM-ER). We have implemented two different approaches for climatological correction: (1) a cumulative density function (CDF) matching approach at a daily timescale and (2) a monthly correction approach. Specifically, the implemented CDF matching approach is a 5th-order polynomial correction as described in Brocca et al. (2011) for matching the CDF of estimated rainfall with respect to reference rainfall, in which the CDF values are computed over the whole calibration period at a daily timescale. The monthly correction approach computes the monthly ratios between the climatology of estimated and reference rainfall, i.e. 12 correction factors per pixel. Then, the SM2RAIN-estimated rainfall is multiplied for the monthly correction factors to obtain the climatologically corrected SM2RAIN-estimated rainfall.
For the global assessment of satellite, reanalysis, and gauge-based rainfall
products we have used the triple-collocation technique. TC can
theoretically provide error and correlations of three products (a triplet)
given that each of the three products is afflicted by mutually independent
errors. Therefore, in principle, TC can be used for assessing the quality of
satellite products without using ground observations (Massari et
al., 2017a). In this study, we have implemented the same procedure as
described in Massari et al. (2017a), i.e. by implementing an
additive error model at a daily timescale, and we refer the reader to this
study for the analytical details. In synthesis, by using the extended TC
method firstly proposed by McColl et al. (2014), it is possible
to estimate the temporal correlation,
Performance of two different configurations at 1009
points in terms of Pearson's correlation,
Performance at 1009 points in terms of Pearson's
correlation,
Several metrics have been used to assess the product performance during the
validation period. As continuous scores we have computed the Pearson's
correlation coefficient (
The results are split in three parts: (1) selection of the optimal configuration of SM2RAIN through the assessment at 1009 points, (2) generation of the global SM2RAIN–ASCAT rainfall data record, and (3) regional assessment of the SM2RAIN–ASCAT data with gauge-based rainfall datasets and global assessments through TC.
As a first step we have collocated satellite soil moisture data from ASCAT
soil moisture H113 and H114, ground-based rainfall observations, and actual
evapotranspiration data from ERA5 in space and time at 1009 points. We have
selected 1009 uniformly distributed points over a regular grid with a spacing
of 1.5
The first test has been dedicated to the filtering of soil moisture data by
using three approaches: (1) SWI, i.e. the REF configuration, (2) SWI with
The second test has been performed on the SM2RAIN equation by using
different drainage functions (VGEN and MUA configurations), by adding the
evapotranspiration component (EVAP) and considering the variability of the
sensing depth,
The final test has been done by applying the daily CDF matching, BC-CDF, and
monthly correction factors, BC-MON, for correcting the climatological bias
in the SM2RAIN-derived rainfall estimates; results are shown in columns 8 and 9
of Fig. 3. For these two configurations, the
improvements with respect to REF are evident but with different magnitude
for the different scores. The BC-CDF significantly improves, while STDRATIO, TS, and FAR show a slight deterioration in
Figure 4 shows the time series of rainfall averaged over the
four regions as obtained from ground observations and from the BC-MON
configuration. The agreement of spatially averaged rainfall with
observations is high with
Time series of mean areal rainfall for the four regions
for observed data, OBS, and the SM2RAIN–ASCAT data record, BC-MON configuration
(
Pearson's correlation,
Regional assessment of the SM2RAIN–ASCAT rainfall data record
and comparison with GPCC, ERA5 and GPM-ER rainfall products. As a reference
the high-quality ground-based datasets in Italy, the United States, India, and
Australia are used. Results in terms of Pearson's correlation,
Global triple-collocation, TC, results.
Based on the tests summarized in the previous paragraph, we have selected the
best configuration using the SWI-Tvar for filtering, Brooks–Corey function for
losses, and monthly correction approach for climatological correction.
The addition of an evapotranspiration component, even though it shows some
improvements, has been not used in view of an operational implementation of
the method. The monthly correction approach has been selected as
Best-performing product based on the results of the triple-collocation data shown in Fig. 7. SM2RAIN–ASCAT is performing the best among the three products in Africa, South America, the central western United States, and central Asia, while GPCC is performing the best in the remaining parts of the Northern Hemisphere and in Australia. GPM-ER is the best product in tropical and equatorial regions.
The selected configuration has been applied on a global scale to 839 826 points over which ASCAT soil moisture observations are available. As
a reference dataset for the calibration of the parameter values of the
pre-processing (filtering) of SM2RAIN and of the post-processing, the ERA5
rainfall has been used mainly because of its higher spatial resolution
compared to the GPCC (36 km versus 100 km). However, we have also tested the use
of the two datasets for calibration at 20 000 randomly chosen points, which
showed that the estimated rainfall in the two calibration tests is very
similar. For instance, the median
The SM2RAIN–ASCAT data record so obtained has a spatial sampling of 12.5 km,
a daily temporal resolution, and it covers the 12-year period 2007–2018.
Figure 5 shows
By using all the pixels included in the four regions (Italy, the United States,
India, and Australia), for a total of 29 843 points, the new SM2RAIN–ASCAT
rainfall data record has been compared with reference rainfall observations
in Fig. 6 by considering the whole period 2007–2018.
Specifically, the box plots of different performance metrics (the same of
Fig. 3) are shown and compared with the results obtained
through the GPCC, ERA5, and GPM-ER. By focusing on the SM2RAIN–ASCAT data record
performance over the different regions, it shows better performance in Italy
(median
On a global scale, the TC approach has been implemented by using the triplet
SM2RAIN–ASCAT, GPM-ER, and GPCC by considering the common period 2015–2018
and a daily timescale. In TC analysis we have purposely not considered ERA5 in order to avoid any dependency between the products. Theoretically, the
extended TC approach provides the correlation,
The box plots of
The SM2RAIN–ASCAT data record is freely available at
In this study, we have described the development of the new SM2RAIN–ASCAT rainfall data record highlighting the steps carried out for improving the retrieval algorithm and the pre- and post-processing of the data. The major novelties of the SM2RAIN–ASCAT rainfall data record developed here with respect to previous versions are (1) the application of SM2RAIN at full spatial resolution thus providing a gridded data record with a spatial sampling of 12.5 km, (2) improved sampling and filtering of ASCAT soil moisture data, (3) the application of monthly climatological correction, and (4) the improved calibration strategy.
The SM2RAIN–ASCAT data record has been preliminary assessed at regional
(Figs. 4 and 6) and global (Figs. 5, 7,
and 8) scales in terms of different performance metrics with a larger
emphasis on the temporal correlation,
In the near future, we are going to develop the near-real-time version of the SM2RAIN–ASCAT rainfall product that can be used as an input for applications such as flood prediction (similarly to Camici et al., 2018 and Massari et al., 2018), landslide prediction (Brunetti et al., 2018), and novel applications for agriculture and water resource management.
Mean daily revisit time (d) of ASCAT soil moisture
observations for the period 2007–2012 (only MetOp-A,
Number of days in which ASCAT soil moisture observations
are close to saturation (
Equations used for the performance scores. For the
continuous scores,
All authors contributed extensively to the work presented in this paper. LB conceived of and designed the paper, developed the SM2RAIN code, and performed some of the analysis. PF wrote most of the code and performed most of the computations. LC, CM, SC, SH, and WW contributed to the processing of the input–output datasets, the analysis and exploration of the data, and the preparation and discussion of the results. LS and BB contributed extensively to define the concept of the paper and to define the procedure to be implemented. All co-authors contributed to the editing of the manuscript and to the discussion and interpretation of the results.
The authors declare that they have no conflict of interest.
The authors gratefully acknowledge support from EUMETSAT through the Global SM2RAIN project (contract no. EUM/CO/17/4600001981/BBo) and the “Satellite Application Facility on Support to Operational Hydrology and Water Management (H SAF)” CDOP 3 (grant no. EUM/C/85/16/DOC/15).
This research has been supported by EUMETSAT (Global SM2RAIN project grant no. EUM/CO/17/4600001981/BBo and “Satellite Application Facility on Support to Operational Hydrology and Water Management (H SAF)” CDOP 3 grant no. EUM/C/85/16/DOC/15).
This paper was edited by Alexander Gelfan and reviewed by three anonymous referees.