ArticlePDF Available

Abstract and Figures

This paper presents the first experimental results of a study on the ingestion in the Weather Research and Forecasting (WRF) model, of Sentinel satellites and Global Navigation Satellite Systems (GNSS) derived products. The experiments concern a flash-floodevent occurred in Tuscany (Central Italy) in September 2017. The rationale is that numerical weather prediction (NWP) models are presently able to produce forecasts with a km scale spatial resolution, but the poor knowledge of the initial state of the atmosphere may imply an inaccurate simulation of the weather phenomena. Hence, to fully exploit the advances in numerical weather modelling, it is necessary to feed them with high spatiotemporal resolution information over the surface boundary and the atmospheric column. In this context, the Copernicus Sentinel satellites represent an important source of data, because they can provide a set of high-resolution observations of physical variables (e.g. soil moisture, land/sea surface temperature, wind speed) used in NWP models runs. The possible availability of a spatially dense network of GNSS stations is also exploited to assimilate water vapour content. Results show that the assimilation of Sentinel-1 derived wind field and GNSS-derivedwater vapour data produce the most positive effects on the performance of the forecast.
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
European Journal of Remote Sensing
ISSN: (Print) 2279-7254 (Online) Journal homepage:
Effect of the ingestion in the WRF model of
different Sentinel-derived and GNSS-derived
products: analysis of the forecasts of a high impact
weather event
Martina Lagasio, Luca Pulvirenti, Antonio Parodi, Giorgio Boni, Nazzareno
Pierdicca, Giovanna Venuti, Eugenio Realini, Giulio Tagliaferro, Stefano
Barindelli & Bjorn Rommen
To cite this article: Martina Lagasio, Luca Pulvirenti, Antonio Parodi, Giorgio Boni, Nazzareno
Pierdicca, Giovanna Venuti, Eugenio Realini, Giulio Tagliaferro, Stefano Barindelli & Bjorn Rommen
(2019): Effect of the ingestion in the WRF model of different Sentinel-derived and GNSS-derived
products: analysis of the forecasts of a high impact weather event, European Journal of Remote
Sensing, DOI: 10.1080/22797254.2019.1642799
To link to this article:
© 2019 The Author(s). Published by Informa
UK Limited, trading as Taylor & Francis
Published online: 17 Jul 2019.
Submit your article to this journal
View Crossmark data
Eect of the ingestion in the WRF model of dierent Sentinel-derived and
GNSS-derived products: analysis of the forecasts of a high impact weather
Martina Lagasio
, Luca Pulvirenti
, Antonio Parodi
, Giorgio Boni
, Nazzareno Pierdicca
, Giovanna Venuti
Eugenio Realini
, Giulio Tagliaferro
, Stefano Barindelli
and Bjorn Rommen
CIMA Research Foundation, Savona, Italy;
DICCA, Department of Civil, Chemical and Environmental Engineering, University of Genoa,
Genoa, Italy;
Department of Information Engineering, Electronics and Telecommunications, Sapienza University of Rome, Rome, Italy;
Department of Civil and Environmental Engineering, Politecnico di Milano, Milan, Italy;
Geomatics Research & Development srl,
Lomazzo, Italy;
European Space Agency (ESA-ESTEC), Noordwijk, The Netherlands
This paper presents the rst experimental results of a study on the ingestion in the Weather Research
and Forecasting (WRF) model, of Sentinel satellites and Global Navigation Satellite Systems (GNSS)
derived products. The experiments concern a ash-oodevent occurred in Tuscany (Central Italy) in
September 2017. The rationale is that numerical weather prediction (NWP) models are presently
able to produce forecasts with a km scale spatial resolution, but the poor knowledge of the initial
state of the atmosphere may imply an inaccurate simulation of the weather phenomena. Hence, to
fully exploit the advances in numerical weather modelling, it is necessary to feed them with high
spatiotemporal resolution information over the surface boundary and the atmospheric column. In
this context, the Copernicus Sentinel satellites represent an important source of data, because they
can provide a set of high-resolution observations of physical variables (e.g. soil moisture, land/sea
surface temperature, wind speed) used in NWP models runs. The possible availability of a spatially
dense network of GNSS stations is also exploited to assimilate water vapour content. Results show
that the assimilation of Sentinel-1 derived wind eld and GNSS-derivedwater vapour data produce
the most positive eects on the performance of the forecast.
Received 24 December 2018
Revised 18 April 2019
Accepted 9 July 2019
Numerical weather model;
Sentinel-1; Sentinel-3; data
The progresses achieved in numerical weather prediction
(NWP) presently allow the models to produce forecasts
with a grid spacing of the order of 1 km (the so-called
cloud-resolving grid spacing) (e.g. Cardoso, Soares,
Miranda, & Belo-Pereira, 2013). High-resolution simula-
tions are required especially for complex topography
areas (e.g. Mediterranean Sea coastline), even though
they are very expensive from a computational point of
view, because some processes, such as strong convection,
can only be captured when the resolution of the numer-
ical model is higher than 2 km (Iriza, Dumitrache,
Lupascu, & Stefan, 2016). However, high-resolution
NWP models are generally fed by low-resolution and/or
not timely updated data and this implies a poor knowl-
edge of the initial state of both atmosphere and surface at
small scales. The uncertainties in the high-resolution
representation of the surface and the atmosphere may
result in an inaccurate simulation of the severe weather
phenomena in terms of timing, location and intensity.
The reliability of initial and boundary data may have
a large impact on NWP simulations, even larger than
that of the model grid spacing (Schepanski, Knippertz,
Fiedler, Timouk, & Demarty, 2015).
At present, NWP models operating at cloud-
resolving grid spacing can take advantage of the avail-
ability of several free of charge Earth Observation (EO)
data characterized by high spatial and/or temporal reso-
lution. In particular, the Copernicus Sentinel satellites are
able to provide information on the surface boundary and,
through the Synthetic Aperture Radar (SAR) interfero-
metry (InSAR) technique, on the atmospheric column
(Mateus, Catalao, & Nico, 2017), with a spatial resolution
consistent with that of cloud-resolving models (1 km).
Very high temporal resolution data about atmospheric
water vapour can be also derived from GNSS (Global
Navigation Satellite Systems) data (Bevis et al., 1994),
although they are basically point measurements. It can
be expected that ingesting products derived from the
aforementioned EO data into NWP models might sig-
nicantly reduce weather forecast uncertainties.
However, while some investigations on the assimilation
in NWP models of low resolution (tens of km) EO-
derived products (e.g. soil moisture extracted from the
Soil Moisture and Ocean Salinity mission data), are avail-
able in the literature (e.g. Muñoz Sabater, Fouilloux, & De
CONTACT Luca Pulvirenti CIMA Research Foundation, via Armando Magliotto 2, Savona 17100, Italy
© 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Rosnay, 2012), few studies were conducted on the inges-
tion of high-resolution EO products. Thus, the European
Space Agency (ESA) issued an intention to tender to
investigate whether the latter kind of products could be
useful to predict, at the high spatiotemporal resolution,
the atmospheric phenomena that lead to extreme events
and intense atmospheric turbulence phenomena.
This paper presents the rst results of a set of
experiments carried out in the framework of the
STEAM (SaTellite Earth observation for
Atmospheric Modelling) research project, whose
objective was to answer to the aforementioned ESA
specic request. To pursue this goal, a NWP model,
namely the Weather Research and Forecasting (WRF)
model, was used at cloud-resolving scale to analyse
the impact of the ingestion of dierent Sentinel-
derived surface observations and of the integrated
precipitable water vapour derived from GNSS
To the best of our knowledge, only a few papers on the
assimilation in NWP prediction models of water vapour
maps derived from SAR data by applying the InSAR
technique are available in the literature. Mateus et al.
(2018) used a horizontal grid spacing of the order of the
cloud permitting limit (3 km), while Mateus, Tomé, Nico,
Member, and Catalão (2016) and Pichelli et al. (2015)
worked at the cloud-resolving limit (1 km), but with
a quite small spatial domain (200 × 200 km
203 km
, respectively). Furthermore, Panegrossi, Ferretti,
Pulvirenti, and Pierdicca (2011)presentedarst attempt
to establish whether high-resolution SAR-derived soil
moisture data might be useful for operational NWP
applications; they used the Fifth Generation NCAR/
Penn State Mesoscale Model (MM5) at 1 km grid spacing
but, again, over a small domain (120 × 120 km
). Data
about the sea and land surfaces derived from Sentinel
data were never used for weather forecast applications
so far.
With respect to the aforementioned papers, in this
study a bigger spatial domain (1500 × 1500 km
considered, still working at the cloud-resolving limit.
Moreover, not only one variable, but an ensemble of
variables (derived from GNSS and Sentinel data), are
ingested. The ingestion of multiple high spatial and/
or temporal resolution EO products, together with
GNSS (Global Navigation Satellite System) data, into
a cloud-resolving NWP model, operated at 1.5 km
grid spacing over a wide region, represent, therefore,
the key novel contribution brought by this study.
Several model runs have to be accomplished to pro-
duce a complete set of results and each high-resolution
run is very expensive from a computational point of view.
Consequently, the number of case studies analysed
throughout the whole duration of the STEAM project
(18 months) was necessarily limited to 23events.The
rst set of STEAM experiments, presented in this paper,
were conducted considering a major ash-ood event
occurred in the Livorno town (Tuscany, Central Italy)
in September 2017 with a dramatic deaths toll (9 people),
and severe economic impacts. The experiments included
the control simulation (i.e., no ingestion of any variable in
the WRF model) used as benchmark, the ingestion of one
variable at a time and the ingestion of a combination of
multiple variables (Sentinel-derived products and/or
GNSS observations). The analysis was based on the com-
parison between the forecasted rain rate (cumulated in
a 12-h interval) and the spatially interpolated rain rate
measured by a network of rain gauges.
The paper is organized as follows. The Materials
and Methods section rstly discusses the rationale
behind the choice of the physical variables ingested
in the WRF model and presents the synoptic back-
ground and the meteorological characteristic of the
Livorno case study. The adopted model setup, the
assimilation methods and the validation method are
also described in this section. The Variables retrieval
section outlines the methods adopted to retrieve the
physical variables that are not already available as
high-level products and introduces the data used to
carry out the experiments. The Results and
Discussion section is dedicated to the description of
the experiments and the analysis of their results. The
nal section draws the main conclusions of the study.
Materials and methods
Selection of the variables to be ingested in WRF
To select the variables to be ingested into the WRF
model, the Observing Systems Capability Analysis
and Review (OSCAR) tool, developed by the World
Meteorological Organization (WMO) in support of
Earth Observation applications, studies and global
coordination ( was
adopted. Through OSCAR, a set of requirements for
the observation of meteo-hydrological variables of
interest in WMO programs and dierent application
areas is available. Requirements are expressed for
geophysical variables in terms of the following cri-
teria: A) uncertainty, B) horizontal and (when applic-
able) vertical resolutions, C) observing cycle (i.e., the
temporal resolution), D) timeliness, E) stability
(when applicable). One of the OSCAR application
areas is represented by high-resolution NWP and its
relevant variables are listed in Table 1.
Looking at Table 1, it can be deduced that, in
principle, several variables can be relevant for high-
resolution NWP applications. Then, an analysis was
performed with the aim of selecting a subset of these
variables. The analysis was based on the following
points: I) possibility to estimate the variable using
satellite EO data (in particular Sentinel ones); II)
utility of the variable for the specic case study con-
sidered here (e.g. snow cover, or sea ice were
obviously not considered because the event occurred
in the Mediterranean area during the summer); III)
compliance with the OSCAR requirements.
Regarding point III), for each criterion OSCAR pro-
vides three values: 1) threshold, which is the mini-
mum requirement to be met to ensure that data are
useful; 2) goal, which is an ideal requirement above
which further improvements are not necessary; 3)
breakthrough, which is an intermediate level between
threshold and goal and that, if achieved, would result
in a signicant improvement for the targeted applica-
tion. Among the criteria listed above, the priority was
given to horizontal resolution, in agreement with the
discussion done in the introduction of this paper.
The outcomes of the analysis led us to select the
following variables: Integrated Water Vapour (that
can be retrieved using Sentinel-1 or GNSS data),
Soil Moisture at the surface and Wind Vector over
the surface (that can be retrieved using Sentinel-1
data), Land and Sea Surface Temperature (that can
be retrieved using Sentinel-3 data).
To briey outline the analysis carried out to select
the variables to be ingested in WRF, Soil Moisture
(SM) is taken as an example. Table 2 is again derived
from OSCAR and shows the requirements for SM.
Firstly, the horizontal resolution requirement was
considered and the literature regarding SM retrieval
using SAR (and Sentinel-1 in particular) was exam-
ined. It was found that Sentinel-1 (S1) derived SM
maps are generally produced with a spatial resolution
of the order of 1 km (Balenzano et al., 2013;
Hornacek et al., 2012; Pierdicca, Pulvirenti, & Pace,
2014). It must be pointed out that, although the
nominal S1 resolution (5 × 20 m
single look) could
in principle allow for producing SM maps with
a higher resolution, a multi-look processing using
a large number of looks and/or a smoothing/low
pass lter are generally applied to cope with the
speckle noise characteristic of SAR images.
Nonetheless, a resolution of 1 km enables to full
the OSCAR requirement at the goal level. For what
concerns the uncertainty, it is dicult to provide
a reliable score for SM, because reference (ground
truth) data are available in few geographic areas
(e.g. the ground stations belonging to the
International Soil Moisture Network, or ad-hoc cam-
paigns). In the literature, it was shown that it is
possible to estimate SM with accuracy (root mean
square error RMSE) from 0.04 to 0.08 m
Hajj et al., 2016), thus meeting the OSCAR require-
ments at least at threshold level. However, these
scores often refer to specic test sites, so that at larger
scales (e.g. national scale), higher RMSE can be
expected, especially if SAR observations are
Table 1. Variables relevant for high-resolution NWP according to the WMO-OSCAR database (
Subdomain Variables
Basic Atmospheric Air pressure (at surface) Air specic humidity (at surface) Air temperature (at surface)
Atmospheric temperature Specic humidity Integrated Water Vapour (IWV)
Wind (horizontal) Wind (vertical) Wind speed over the surface (horizontal)
Wind vector over the surface
Clouds and
Accumulated precipitation (over 24 h) Cloud base height Cloud cover
Cloud drop eective radius Cloud ice
Cloud ice (total column) Cloud liquid water (CLW) Cloud liquid water (CLW) total column
Cloud top height Cloud type Precipitation intensity at surface (liquid or
Precipitation intensity at surface (solid) Precipitation type at the surface
Aerosols and radiation Earth surface albedo Fraction of Absorbed PAR (FAPAR) Long-wave Earth surface emissivity
Upward short-wave irradiance at TOA Upward long-wave irradiance at
Ocean Dominant wave direction Dominant wave period Sea surface temperature
Signicant wave height
Land surface Land surface temperature Leaf Area Index (LAI) Normalised Dierence Vegetation Index
Snow cover Snow water equivalent Soil moisture at surface
Atmospheric chemistry O3 O3 (Total column)
Sea Ice Sea-ice cover Sea-ice surface temperature Sea-ice thickness
Table 2. Requirements dened for soil moisture at the surface for high-resolution NWP according to the WMO-OSCAR database
Goal Breakthrough Threshold
Uncertainty 0.02 m
0.04 m
0.08 m
Stability/decade (if applicable)
Horizontal Resolution 1 km 5 km 40 km
Vertical Resolution
Observing Cycle 60 min 3 h 6 h
Timeliness 30 min 60 min 6 h
performed under dense vegetation conditions (e.g.
Hajnsek et al., 2009). In any case, the fullment of
the spatial resolution and, to some extent, accuracy
requirements led us to select SM among the variables
ingested by WRF. It must be however underlined that
the observation cycle can represent a critical aspect.
Indeed even considering S1 ascending and descend-
ing orbits, the observation cycle varies between 12
h (best case) and six days (worst case). Similar lines
of reasoning were followed to select Integrated Water
Vapour (IWV), Wind speed (W
) and direction (W
(i.e., the wind vector), Land Surface Temperature
(LST) and Sea Surface Temperature (SST).
Case study
To comply with the objective of the STEAM project
(discussed in the Introduction), a critical aspect was
the choice of the case studies. It was necessary to
search for high impact weather events (HIWEs)
observed as close as possible to their onset by both
S1 and Sentinel-3 (S3) and occurred in a geographic
area where reliable validation data were available. To
full these very strict requirements, a review of the
HIWEs happened in Italy (where a dense rain gauge
network useful for validation purposes is available) in
recent years (20162018), on the basis of European
Severe Weather Database and the corresponding
Sentinel data availability was carried out. This review
led us to choose the Livorno ash ood as the rst
case study to be analyzed in the framework of the
STEAM project.
Starting from the afternoon-evening of Saturday,
9 September, a large trough deepened on the western
Mediterranean, recalling an intense ow of currents
from the south, mild and extremely humid, on all the
Tyrrhenian sectors and on the part east of the
Ligurian Sea. From the evening of Saturday
9 September, the freshest airow associated with
vorticity at 500 hPa was supportive of instability
conditions on the Tuscany region (not shown). The
environment was also conducive to the development
of intense local convective precipitation systems per-
sistent, not only because of the slow evolution of the
depression area, but also because of the shear of the
winds (variation of the intensity and direction of the
wind along the vertical column) well highlighted by
the deep level shearwhich determined a separation
between the updraft area (rising currents that feed the
storms) and that of downdraft (descending currents
that generate the wind band), favouring locally sta-
tionary storms. However, it is very important to
consider, from a predictability standpoint, how the
entire central Tyrrhenian sea, and large part of cen-
tral Italy were prone to the potential occurrence of
very intense, persistent, and self-regenerating thun-
derstorm phenomena as conrmed by the values of
the Severe Weather Threat Index (SWEAT) that mea-
sures thunderstorm potential by examining low-level
moisture, convective instability, jet maxima, and
warm advection (Figure 1). These conditions resulted
in about 250 mm of rain fell in 2 h between 02:00 and
04:00 UTC of 9/10 September over the Livorno city
area causing the death of nine people.
Available data
For the Livorno event, S1 and S3 data acquired
between the afternoon of September 8 and the eve-
ning of 9 September 2017 were available. In particu-
lar, the following data were directly available through
the Sentinel catalogues: 1) LST and SST derived from
the Sea and Land Surface Temperature Radiometer
(SLSTR) onboard S3, generated on a 1 × 1 km
2) wind vector (W
and W
at 10 m asl) included in
the Ocean Wind Field (OWI) component of the
Sentinel 1 Level-2 Ocean (OCN) product (again gen-
erated on a 1 × 1 km
grid). More specically, for
Figure 1. SWEAT index at 07 UTC of September 10, 2017 (ECMWF 25 km run, 12 UTC September 9, 2017).
SST, the Sentinel 3 data acquired on
9 September 2017 at 20:36 UTC were gathered,
while for LST, the Sentinel 3 data acquired on
9 September 2017 at 09:50 UTC were used. For
what concerns SM, W
and W
, S1 observed the
Livorno area on 8 September 2017 at 17:14 UTC.
To derive SM from S1 data and IWV from GNSS
data we relied on retrieval algorithms (described in
the Variables retrieval section).
WRF model setup and data assimilation methods
The WRF model (Skamarock et al., 2008) was selected
as the numerical weather model to accomplish all the
experiments carried out within the framework of the
STEAM project. It is a compressible non-hydrostatic
model that was developed at the National Center for
Atmospheric Research (NCAR) in collaboration with
several institutes and universities for operational
weather forecasting and atmospheric science research.
For this work, WRF version 3.8.1 was adopted. Three
domains with two-way nesting setup (Figure 2)were
used, at 13.5 (250 × 250 horizontal grid points), 4.5
(451 × 450) and 1.5 km (943 × 883) spatial resolution,
with 50 vertical levels (all domains top reached 50
hPa). The selected WRF model setup was based on
previous results for this kind of events (Fiori et al.,
2014,2017; Lagasio, Parodi, Procopio, Rachidi, &
Fiori, 2017). The two inner domains (4.5 and
1.5 km) allow solving explicitly many convective pro-
cesses (Kain et al., 2008; Kain, Weiss, Levit, Baldwin, &
Bright, 2006) so an explicit treatment of convection is
chosen. As for the 13.5 km domain, a parameterization
scheme closest to the one used in the GFS global
model is chosen (Han & Pan, 2011)
Concerning the surface layer, the MM5 scheme
was adopted; this scheme uses stability functions
from Dyer and Hicks (1970), Paulson (1970), Webb
(1970) to compute surface exchange coecients for
heat, moisture, and momentum. A convective velo-
city following Beljaars (1995) was used to enhance
surface uxes of heat and moisture. The Rapid
Update Cycle (RUC) was chosen for land processes;
the scheme has a multi-level soil model with higher
resolution in the upper soil layer (0, 5, 20, 40, 160,
300 cm is default). The planetary boundary layer
(PBL) scheme is based on the diagnostic non-local
Yonsei University PBL (Hong, Noh, & Dudhia, 2006)
scheme, as the next generation of the Medium-Range
Forecast (MRF) PBL, also using the countergradient
terms to represent uxes due to non-local gradients.
As for the microphysics, the WRF single-moment
6-class (WSM6) scheme was used; it extends the
WSM5 scheme to include graupel and its associated
processes. Of the three WSM schemes, the WSM6
scheme is the most suitable for cloud-resolving
grids, considering the eciency and theoretical back-
grounds (Hong & Lim, 2006). Finally, the radiative
processes were parameterized by means of the long-
wave and shortwave Rapid Radiative Transfer Model
(RRTMG) schemes (Iacono et al., 2008). The initial
and lateral boundary conditions were derived from
NCEP-GFS (National Centers for Environmental
Prediction Global Forecast System) analysis and fore-
cast data available at a horizontal resolution of 0.25°×
0.25° and a time resolution of 3 h.
The data ingestion was performed according to
three dierent methodologies: direct insertion,
a nudging-like technique, and nally a 3DVAR
assimilation. Direct insertion is meant hereafter as
the substitution of a given variable in the NWM elds
with the corresponding one retrieved by EO sensors;
it was applied for SST and LST. The nudging-like
technique was applied for the SM observations; it
included dierent steps. Firstly, a dierence map
between the S1-derived SM (reprojected to the WRF
grid) and the WRF SM from the rst level of soil
model (in correspondence with the points observed
by S1) was produced; then the resulting map was
interpolated to ll the gaps in the S1-derived maps
and added to the original WRF SM eld in the rst
(supercial) level. Finally, the new SM eld was pro-
pagated in the underlying vertical levels by a vertical
prole correction through the linear interpolation of
the dierence between the observed and simulated
SM assuming a dierence equal to 0 in the deepest
level (the S1 observation is only supercial, in the
deepest layers the value of the model is more reli-
able). Figure 3 shows the SM maps resulting from the
procedures described above. It can be noted that in
the lowest layer there is no inuence of the satellite
observation and that this inuence increases in the
upper layers.
Figure 2. WRF nested domains used for the simulations
and W
from Sentnel-1 and GNSS were assimi-
lated using the 3 DVAR technique. In particular, for
GNSS a 3-h cycle technique was implemented (to take
advantage of the high temporal resolution). The 3DVAR
main purpose is to provide an optimal estimate of the
current status of equation (1) (Ide, Courtier, Ghil, &
Lorenc, 1997).
 (1)
where xis the analysis to be found that minimizes the
cost function J(x), x
is the rst guess of the NWP
model, y° is the assimilated observation, and y=H(x)
is the model-derived observation transformed from
the analysis xby the observation operator Hfor
comparison against y°. The solution of (1) represents
a posteriori maximum likelihood (minimum var-
iance) estimate of the true state given two sources
of data: the rst guess x
and the observation y°. The
analysis t to this data is weighted by estimates of
their errors Band R, which denote the background
error covariance matrix and the observation error
covariance matrix, respectively.
The version 3.9.1 of the Data Assimilation system
built within the WRF (WRFDA) was chosen. The
Control Variable option 5 (CV5) of the WRFDA
package was used for the Bmatrix calculation using
the National Meteorological Center (NMC) method
(Wang et al., 2014). The NMC method was applied
over the entire month of October 2013 with a 24-h
lead time for the forecasts starting at 00:00 UTC and
a 12-h lead time for the ones initialized at 12:00 UTC
of the same day. The dierences between the two
forecasts (t+ 24 and t+ 12) valid for the same refer-
ence time were used to calculate the domains specic
error statistics.
The Rmatrix was assumed to be diagonal, as done
in most of the models (Bouttier & Courtier, 1999).
Thus, observation error correlations are often
assumed to be zero, considering distinct measure-
ments aected by physically independent errors.
Conversely, observation error variances are assumed
equal to the instrumental errors, thus varying
between dierent variables.
Validation method
To validate the model output for each experiment the
Method for Object-Based Diagnostic Evaluation
(MODE) tool (Davis, Brown, Bullock, & Halley-
Gotway, 2009) was used. Simply speaking, MODE is
a basic algorithm for image segmentation and object
Figure 3. First row: S1 derived soil moisture; second and third rows: six layers SM maps used as WRF input
matching, specically developed for meteorological
applications. After the identication of the objects in the
precipitation maps (observed and forecasted), MODE
assignstotheobjectaseriesofattributesdened both
geometrically (e.g. size, orientation) and based on the
precipitation values inside the objects. Then, to summar-
ize the performance of forecasts, it compares the attri-
butes between matched forecasted and observed objects.
The use of this validation method is mainly due to the fact
that when comparing high-resolution observational data
analysis and cloud-resolving meteorological forecast,
especially in case of deep moist convective and highly
localized phenomena, the traditional verication meth-
ods suer from the so-called double penaltyissue and
hence alone cannot provide a measure of spatial and
temporal match between the forecast and observed
meteorological patterns. In this respect, it is then prefer-
able to take advantage of features based verication tech-
nique, such as MODE. For sake of clarity and
conciseness, the MODE evaluation in terms of pairs of
objects attributes is hereafter summarized using three
representative output indices: centroid distance, angle
dierence and area ratio (Table 3).
Among the dierent parameters, MODE provides also
some classical statistical scores. Then to assess the perfor-
mance of the forecasts also the probability of detection
(PODY), the false alarm ratio (FAR) and the frequency
bias (FBIAS) were chosen. These parameters were
derived from a contingency table that shows the fre-
quency of yesand norain forecasts and occurrences.
The four combinations of forecasts (yes or no) and
observations (yes or no) are: 1) hit: rain forecasted and
actually occurring; 2) miss: rain not forecasted and actu-
ally occurring; 3) false alarm: rain forecasted and actually
not occurring; 4) correct negative: rain not forecasted and
actually not occurring.
PODY is given by the number of hits divided by
the total number of observed rainy events:
PODY ¼hits
hits þmisses perfect score ¼1ðÞ(2)
FAR is given by the number of false alarms divided
by the total number of forecasted rainy events:
FAR ¼false alarms
hits þfalse alrms perfect score ¼0
FBIAS is given by the total number of forecasted
rainy events divided by the total number of observed
rainy events:
FBIAS ¼hits þfalse alarms
hits þmisses perfect score ¼1ðÞ(4)
Variables retrieval
As previously pointed out, SST and LST were directly
available as Sentinel-3 (S3) level-2 products (currently
through the Sentinel-3 Pre-Operations Data Hub),
while W
and W
were directly available as S1
level-2 products (through the Copernicus Open
Access Hub). Hence, for these variables, there was
no need to use a retrieval algorithm. Conversely, SM
and IWV had to be retrieved. As for the latter vari-
able, the quantity that can be estimated from satellite
data is the Zenith Total Delay (ZTD) from GNSS and
the Slant Total Delay (STD) from S1 using the InSAR
technique. ZTD represents the delay induced by the
troposphere on GNSS signals in the zenith direction,
while STD is the dierence between the slant range
atmospheric delays of the master and the slave images
of an Interferometric pair. Since it was found that, for
the Livorno event, there was little water vapour in the
atmosphere at the S1 acquisition time, only GNSS-
derived ZTD were considered in this study. The
methods used to retrieve SM and ZTD are briey
described hereafter.
Soil moisture retrieval from sentinel-1 data
The algorithm developed to estimate SM was con-
ceived to combine a multi-temporal approach with
a good computational eciency. The latter character-
istic is particularly suitable for an operational appli-
cation, such as the use of SM maps in NWP models.
A multi-temporal approach to estimate SM from SAR
assumes that the temporal scale of variation of soil
roughness is considerably slower than that of SM.
Hence, if a dense time-series of SAR data is available,
as expected using S1, short-term changes in the back-
scattering coecient σ° (that represents the SAR
measurement) are basically related to SM variations.
(Balenzano, Mattia, Satalino, & Davidson, 2011; Kim
et al., 2012; Pierdicca, Pulvirenti, & Bignami, 2010).
The algorithm is based on a multi-temporal max-
imum likelihood (ML) approach (e.g. Kim et al.,
2012) used to invert a direct scattering model of
backscattering from a bare soil, namely that proposed
by Oh (2004). The Oh Model relates the state of the
soil, described by the pair (SM,s), where sdenotes the
height standard deviation of the rough surface, to the
Table 3. Description of the MODE-derived spatial indices used in this work.
Index Description
Centroid Distance Centroid Distance: Provides a quantitative sense of spatial displacement of forecast (Best score 0).
Angle Dierence For non-circular objects, gives measure of orientation errors (Best score 0)
Area Ratio Provides an objective measure of whether there is an over- or under-prediction of areal extent of forecast (Best score 1).
backscattering coecient σ°. This relationship is
represented in the form of lookup table (LUT) (Kim
et al., 2012; Pierdicca, Pulvirenti, Bignami, & Ticconi,
2013; Pierdicca et al., 2014), which was generated by
applying the Oh model considering as inputs 50
values of s(between 0.5 e 4.5 cm with
a discretization of 0.0816 cm), 100 values of SM
(between 0.05 e 0.4 m
with a discretization of
0.0035 m
) and 13 values of incidence angle θ
(between 26° and 50° with a discretization of 2°).
The range of SM and svalues is approximately the
same used by Pierdicca et al. (2013), (2014) and
accounts for the range of validity of the Oh model.
As for the θrange, with respect to the S1
Interferometric Wide Swath (IWS) acquisition mode
nominal incidence angle range, a larger interval was
considered for the LUT to account for topography.
For each pair (SM, s), 13 values of σ° at VV polariza-
tion (one for each θvalue) were computed (indicated
as σ0
VV;model). Note that VV is the default co-polarized
channel of S1.
The retrieval algorithm assumes that a time series
of M+ 1 measurements (at the current time tand
at Mprevious times t1, ..., t M) is available. It
performs a least square search minimizing the square
dierence between measured and modelled values of
d½SMðtÞ;SMðt1Þ; :::; SMðtMÞ;s;θ
VV;soil ðtÞjdB σ0
VV;model ½SMðtÞ;s;θjdB
VV;soil ðt1ÞjdB σ0
VV;model ½SMðt1Þ;s;θjdB
þ::: ::: þσ0
VV;soil ðtMÞjdB σ0
VV;model ½SMðtMÞ;s;θjdB
where dis the cost function that has to be minimized
and the symbol jdB indicates that the backscatter
values are expressed in logarithmic units. In (5),
VV;soil refers to the soil contribution to the S1 mea-
surements, so that, if vegetation is present, its eects
on the S1 backscattering measurements must be cor-
rected before determining the cost function d. The
choice of Mis a compromise between SM estimation
accuracy and computational time; M= 4 was chosen
as done by Pierdicca et al. (2014).
The correction of the vegetation eects on σ°is
a very critical point for any SM retrieval approach. In
fact it is usually based on semi-empirical models that
may lack generality. For this study, the Water Cloud
Model (WCM, Attema & Ulaby, 1978) was used, with
the parameters proposed by Bindlish and Barros
(2001) for all land uses. In the WCM, the vegetation
is simply represented by a bulkparameter, such as
the vegetation water content W. Previous studies (e.g.
Jackson et al., 2004,1999) have shown that there
exists a direct relationships between Wand the
Normalized Dierence Vegetation Index (NDVI),
which can be derived from optical data such as,
Sentinel-2, Landsat, or MODIS ones. Here the
empirical relationship found by Pierdicca et al.
(2013) was used (and NDVI data were extracted
from the MOD13Q1 MODIS product, see Huete
Didan, van Leeuwen, Miura, & Glenn, 2011):
W¼11:92 NDVI 2:73 kg=m2
Note that if W< 0.25 kg/m
, the inuence of vegeta-
tion on the radar signal is considered as negligible
and the S1 backscatter is not corrected (Pierdicca
et al., 2013). Instead, if W>5 kg/m
the retrieval of
SM is assumed to be totally insensitive to SM, so that
pixels are masked in this case.
Zenith total delay retrieval from GNSS data
Water vapour content of the atmosphere low layer
(up to about 10 km), known as troposphere or neu-
tral atmosphere, aects GNSS signals by changing
their propagation velocities with respect to that of
the light in vacuum. A diminished speed results in
a larger observed time Δtto cover the satellite-
receiver distance, and eventually to a larger satellite-
receiver observed distance SobscΔt. The path delay in
length (Slant Total Delay STD) can be therefore
dened as follows (here we disregard the ray
r¼Sobs S¼òs
ds ¼òsnsðÞ1ðÞds
where sis the actual signal path here considered equal
to the straight satellite-receiver line, nis the refrac-
tion index and N106n1ðÞis the refractivity. By
assuming a horizontally layered water vapour distri-
bution, one can express such slant delay as a linear
function of the delay along the vertical direction, i.e.,
the ZTD, above each receiver:
r¼ZTDrmf (8)
where mf is the known mapping function, that pro-
jects the unknown zenith delay onto the satellite-
receiver slant path. Most of the modern mapping
functions are expressed using a continued fraction
in terms of 1/sin(ε) (Marini, 1972), εbeing the satel-
lite elevation, truncated after the third term (Niell,
mf εðÞ¼ 1
sin εðÞþ a
sin εðÞþ b
sin εðÞþc
The a;band cparameters are usually taken from
a global model, the most used being Niell
(Niell,1996), GMF (Böhm, Niell, Tregoning, &
Schuh, 2006) and VMF (Böhm & Schuh, 2004). The
parameters of the VMF (Vienna Mapping Function)
are estimated from ECMWF numerical weather
model (Kouba, 2008) with a temporal resolution of
6 h and a spatial resolution of 2°in longitude and 2.5°
in latitude.
It is worth reminding here that the tropospheric
delay (the word delay is usually referred to the extra
distance and is expressed in meters) due to water
vapour, is just one out of many other systematic
errors aecting GNSS observations which are to be
accounted for in order to achieve sub-centimetre
accuracy positions. Two dierent strategies are cur-
rently used to deal with those systematic eects: one
is based on a combination of the observations of two
receivers to a satellite pair, allowing for the removal
of the common systematic eects, the other models
most of the errors and relies on the currently
achieved high accuracy of satellites ephemerides
(Héroux & Kouba, 2001). This last strategy, known
as Precise Point Positioning (PPP) has been adopted
for the STEAM experiment.
PPP mainly models each satellite-receiver observed
distance as a function of the unknown station coordi-
nates, receiver clock oset, ZTD (a proper combination of
dual frequency data, iono-free combination, gets rid of
the ionospheric contribution) and the phase ambiguities
(Héroux & Kouba, 2001). A lot of eort has been put in
the modelling of the systematic eect, nowadays
accounted for by all software implementing GNSS data
adjustment for high accuracy positioning purposes.
The observations collected by the GNSS station
from all the satellites simultaneously in view at
a given epoch involve the same ZTD parameter
and receiver clock, while the phase ambiguities
and the coordinates of a xed station are common
to all the dierent epochs of a daily session. The
parameters can be retrieved by applying either
aKalmanlter approach, thus updating their esti-
mates epoch by epoch based on the past observa-
tions only, or a joint least squares adjustment of all
the session observations. In the STEAM experi-
ment, the free and open source software goGPS
(Realini & Reguzzoni, 2013) was adopted. It was
developed by GReD srl, implementing a PPP strat-
egy with a multi-epoch joint least squares adjust-
ment of the observations. The processing settings
are summarized in Table 4.
Results and discussion
The experiments
The WRF simulations were started at the
synoptic hour temporally closest to the time of the
S1 acquisition, i.e., on 8 September 2017 at 18:00
UTC and lasted 48 h (until 10 September 2017 at
18:00 UTC).
To apply the multi-temporal SM retrieval algo-
rithm previously described, not only the S1 data
acquired on 8 September 2017, but also those
acquired on September 2, August 27, August 21 and
15 August 2017 were used and the ground range
detected (GRD) products were chosen. The GRD
data were multi-looked (10 × 10), calibrated and
geocoded using the 30 m SRTM DEM. All the
Sentinel-derived data were resampled and reprojected
to a reference WRF grid (that of the inner domain:
Geographic Lat/Lon, WGS84, ~1.5 km of pixel size,
corresponding to 0.0135°). The nearest neighbour
approach was applied for this purpose, except for
the SM maps for which a pixel aggregate resampling,
which averages all the pixel values that contribute to
the output pixel, was adopted.
ZTD time series at 30-s resolution were estimated
for all the geodetic permanent stations within the
WRF domain providing free raw GNSS observations.
For the selected event, the data of 375 stations were
available. An accuracy assessment was performed by
comparing the GNSS-derived ZTD to the eight Italian
radiosounding datasets, yielding a mean dierence of
3 mm and a standard deviation of 14.6 mm (a total of
43 radiosonde launches has been considered).
Firstly, the sensitivity of the WRF model to each
single variable was analysed, then the ingestion of one
variable at a time was accomplished (besides the
control simulation run, as underlined in the
Introduction). The values of SM and wind eld
were assumed as acquired at 18:00 UTC (although
S1 acquisition time was 17:40), i.e., at the starting
time of the simulations.
S3 derived SST and LST data were ingested into
the model via a direct insertion, as previously pointed
out. For this purpose, the WRF simulation was
stopped at the acquisition time of each variable and
then restarted after having replaced the WRF derived
variable with the S3 derived one in a warm restart
mode. Concerning GNSS-derived ZTD, it was assimi-
lated in WRF using the 3DVAR technique. In parti-
cular, the assimilation was cyclically performed every
3 h (throughout the 48h-simulation) in two dierent
ways: 1) considering either a GNSS observation
every minute (in a 30-min time window centred at
the analysis time), 2) considering only the GNSS
Table 4. GPS processing settings.
Satellite elevation cut-o:7 degrees
ZTD random walk regularization: 0.015 m/ ffiffi
ZTD gradients random walk
0.0015 m/ ffiffiffi
Geophysical corrections: IERS 2010 convention
Ambiguity: Float
Mapping function: VMF gridded mapping function
Orbits and satellite clocks: International GNSS Service (IGS)
nal products
observation temporally closest to the analysis time.
The assimilation conguration for each experiment is
summarized in Table 5.
Analysis of the results of the experiments
As validation data, the rain rate measurements pro-
vided by the rain gauge network managed by the
Italian Department of Civil Protection (DPC) were
chosen. It includes more than 5000 rain gauges cover-
ing all the Italian territory. The rain rate measure-
ments were spatially interpolated to be compared
with the model forecasts. To evaluate the forecast
performances, the rain rate cumulated between
September 9 at 18:00 UTC and September 10 at
06:00 UTC (time period when most of the rain fell)
was considered, as the sub-daily period including the
most intense part of the observed ash-ood-
producing storm. Two dierent thresholds on the
cumulated rain (THR hereafter) were considered to
identify the objects in the forecast and observation
maps by using the MODE tool, namely THR 48 mm
and THR 72 mm. Figure 4 and Figure 5 show the
values of the scores computed through MODE
obtained in the experiments of ingestion of one vari-
able at a time
In terms of pairs of objects indices (Figure 4), an
improvement in terms of the centroid distance
(CENTR_DIST) with respect to the control run
(CTRL) is achieved through the assimilation of SST,
SM and the wind eld (W
and W
) for THR72mm
(the assimilation of W
and W
improves the result
for THR>48mm too). Furthermore, the assimilation
of ZTD decreases the angle dierence
(ANGLE_DIFF) for both the thresholds. Finally,
almost all the performed simulations show an
improvement in terms of AREA_RATIO with respect
to CTRL forecast.
Looking at Figure 5, in terms of classical statistical
indices, it can be seen that the ingestion of LST
implied an improvement of the forecast performances
for all the three parameters considered in this analysis
(increase of FBIAS and PODY, decrease of FAR).
However, this improvement did not turn out to be
very signicant except for FBIAS and THR of 48 mm.
Even the ingestion of SST did not produce signicant
impacts on the WRF forecasts. As for SM, its inges-
tion led to an improvement of all the indices except
FBIAS for a threshold of 48 mm. It must be under-
lined that by ingesting SM, the lowest values of FAR
were obtained for both the thresholds.
The most positive impact on the forecast perfor-
mances was obtained by assimilating the wind eld
and W
). In this case, despite of a slight increase
of FAR, a FBIAS close to the perfect score (1, see
Figure 5) was obtained and also PODY improved
with respect to the CTRL. Furthermore, in terms of
rainfall pattern location, the assimilation of the wind
eld-produced improvements in terms of
CENTR_DIST and AREA_RATIO for both the THR
(Figure 4). In this case, the role of high-resolution
observation assimilation seems to be very important.
In fact, the assimilation of high-resolution wind eld
correcting the original WRF wind eld at initializa-
tion time that was forced by the low resolution
(~25 km) GFS data allowed the model to evolve
more in agreement with the physics of the observed
convective system. Figure 6 shows the wind eld
extracted from the S1 OWI level-2 product
(8 September 201817:40 UTC).
The high-resolution observation assimilation at the
initialization time helped the forecast because quanti-
tative precipitation NWP models predictive ability
challenges can derive from the poor knowledge of
the initial state of the atmosphere at small scales lead-
ing to an inevitable model spin-up that often results in
an inaccurate simulation of the convective system in
terms of timing, location and intensity (Sugimoto,
Crook, Sun, Xiao, & Barker, 2009). Thus, the capability
to capture small-scale phenomenon is likely the reason
of the good impact of the S1-derived wind eld on the
forecast, even though these data were assimilated more
than 24 h before the onset of the event. Usually the
data assimilation impacts more the rst 12 h; however,
this is not the rst time in literature that for some
modelling and assimilation setup the impact of obser-
vations assimilation lasts more than 12 h. Koizumi,
Ishikawa, and Tsuyuki (2005) found a positive impact
of rain assimilation on the precipitation up to 18 h;
Dash, Sahu, and Sahu (2013) found out that in a set of
Table 5. Experiment setup summary.
Experiment Assimilation timing Assimilation methods
LST 10 UTC 09/10/2017 Direct Insertion
SST 21 UTC 09/10/2017 Direct Insertion
SM 18 UTC 08/10/2017 Nudging-like
WIND 18 UTC 08/10/2017 3dvar
ZTD3h 3 hour cycling starting at 18 UTC 08/10/2017 3h cycling 3dvar considering GNSS observation every minute in a 30
minutes time window centered at the analysis time
ZTD3h_1ist 3 hour cycling starting at 18 UTC 08/10/2017 3h cycling 3dvar considering GNSS observation temporally closest to the
analysis time
Assimilation at the acquisition time of SM and WIND with
GNSS-ZTD with a 3h cycling 3dvar
3dvar + nudging-like
CTRL No assimilation No assimilation
experiments, 3DVAR improvements in simulation can
be signicant up to 24 h and that the impact of
assimilation is visible up to 72-h simulation, although
with less magnitude. In this case, the choice of the
domain setup (two-way nested with outer domains
boundary very far from the inner one and updated
every 3 h) could help the assimilation impacts to last
more than the usual time range.
To further investigate the impact of wind eld
assimilation at the model initialization, a reference
time step at 02UTC on 10 September 2017, corre-
sponding to the most intense phase of the observed
event, was considered. Then, using the VAPOR
(Visualization and Analysis Platform for Ocean,
Atmosphere, and Solar Researchers, www.vapor. software, the atmospheric ow eld was
analysed compared to the CTRL experiment
(Figure 7). 3D isosurfaces (5 × 10
kg/kg) for the
rainwater, snow and graupel variables have been
rendered in combination with the wind eld at
10 m in case of the CTRL run (panel a) and the
WRF run with the assimilation of the wind eld
(panel c) at the same time instant (02 UTC of
10 September). By inspecting the 10-min horizontal
wind eld, it is possible to argue that the assimila-
tion of the S-1 derived wind eld (Figure 6)con-
tributed to improve, in comparison to the control
run, the persistence and intensity of the conver-
gence responsible for the triggering of the observed
highly-precipitating back-building Mesoscale
Convective System (MCS). Consequently, the pre-
diction obtained by assimilating the wind eld was
denitely more in agreement with the physics of
the observed MCS. Furthermore, while reectivity
Figure 4. Results of the comparison between reference and WRF-derived cumulated rain rate for the Livorno event obtained
using the MODE tool. CENTR_DIST, ANG_DIFF and AREA_RATIO are considered. Left panels: THR 48 mm; right panels: THR72
mm. Blue bars: control run; red bars: ingestion/assimilation of: LST (1st row), SM (2nd row), SST (3rd row), WS and WD (4th row),
ZTD using a GNSS observation every minute; green bar (last row): assimilation of ZTD using only the GNSS observation
temporally closest to the assimilation time (last row). The dark blue lines indicate the perfect scores (0 for CENTR_DIST and
is nearly absent nearby the Tuscany coastlines
(Livorno area) for the OL experiment (panel b),
conversely stronger activity is apparent in panel
b over the Livorno area and downshear the main
convective system.
From Figure 4 and Figure 5, it can be deduced that
the assimilation of the ZTD derived from a GNSS
observation every minute (ZTD3h red bars in the
last row of the gures) and that of the ZTD obtained
using only the GNSS observation temporally closest
to the analysis time (ZTD3h_1ist, green bars) gave
rise to quite similar results. A signicant improve-
ment of FBIAS was achieved with both forecasts,
although associated with a slight increase of FAR
(slightly lower in the ZTD3h_1ist simulation).
PODY slightly improved with respect to the control
simulation (CTRL) only for the ZTD3h simulation. In
terms of spatial pattern location, the CENTROID_
DIST evidences a worsening in the rainfall location
with respect to CTRL simulation, but big improve-
ments were achieved in terms of AREA_RATIO and
ANGLE_DIFF of the convective structure.
From the previous analysis, it emerges that the
WRF forecasts showed a large sensitivity to the
assimilation of the wind eld and water vapour.
For both the thresholds used to identify objects
through the MODE tool, the assimilation of the
wind eld produced the best performances for
FBIAS, PODY and AREA_RATIO, the assimilation
of ZTD produced the best results for
ANGLE_DIFF, while the best performances for
what concerns FAR were obtained by initializing
the WRF run with the S1-derived SM,aspreviously
pointed out. Hence, a further experiment was
Figure 5. Same as Figure 4, but considering FBIAS (perfect score: 1), PODY (perfect score: 1) and FAR (perfect score: 0)
carried out performing the combined assimilation
of the wind eld and the ZTD (derived from
a GNSS observation closest to the analysis time)
This experiment will be synthetically denoted as
WIND+SM+ZTD hereafter. The results are shown
in Figure 8, where the scores obtained in this
experiment are compared with those obtained in
the CTRL run, as previously done, as well as with
the best scores (e.g., highest FBIAS and PODY and
obtained in the previous experiments of ingestion/
assimilation of one variable at a time. It can be
seen that this experiment of combined assimilation
of WIND, ZTD and SM enabled to improve the
performances obtained in the previous experiments
except for the FAR for a THR of 48 mm and the
experiment produced better results with respect to
the experiments of ingestion of a single variable.
Figure 6. Wind eld estimated from the Sentinel-1 observation performed on September 8, 2018 at 17:40 UTC and included in
the OWI level-2 product
Figure 7. 3D simulated structure composed by rainwater (cyan), graupel (yellow) and snow (grey) microphysics species
respectively for the CTRL run (a) and the run with the assimilation of the wind eld (c); the horizontal 10m wind intensity
represented by red vectors. The red line in panels a and c indicates the location of the vertical section of the two structures to
investigate the reectivity values in the middle of the observed convective structure reported in panels b (for CTRL) and d (for
Figure 8. Same as Figure 4 (upper panel) and Figure 5 (lower panel), but considering the combined ingestion of WS and WD,
ZTD using a GNSS observation every minute and SM (red bars); green bars: best results obtained with the ingestion of a single
Figure 9. Panel a: Tuscany region hydro-meteorological alert areas; panel b: areas of interest (S3 in green and A6 in red) for this
event; panels c and d: hourly mean rainfall over the S3 and A6 alert areas, respectively, comparing the observed rainfall (OBS),
the CTRL run, the WIND+SM+ZTD run and a control simulation initialized at 00 UTC on September 9, 2017 (CTRL_00UTC).
In view of an operational forecasting framework,
hourly rainfall forecast produced by the WIND+SM
+ZTD experiment and the CTRL simulation were
compared with the corresponding hourly observed
data. To pave the way to this analysis, it is worth to
notice that the Italian Civil Protection Department
manages the hydro-meteorological warning in terms
of alert areas. The Tuscany region is divided into 26
Alert Areas (Figure 9, panel a), but for this event,
only two areas were largely aected (A6 and S3 high-
lighted in Figure 9, panel b). The hourly rainfall
obtained for these areas is shown in panels c and
d of Figure 9. In case of the S3 area, the WIND+SM
+ZTD simulation produced an improvement in terms
of rainfall peaks (e.g., underestimation of the second
peak is reduced from about 20 mm to about 7 mm)
with respect to the CTRL simulation (panel c), while
the improvement is less pronounced over the A6 area
(panel d). More specically, concerning the A6 area,
the WIND+SM+ZTD simulation enabled
a quantitative precipitation forecast (QPF) improve-
ment for the rst rainfall peak, but without succeed-
ing in mitigating the underestimation of the second
peak. Thus, despite the cyclic update of the forecast
every 3 h with ZTD observations, the forecast is not
able to fully capture the convective system dynamic.
This can be due to the initialization time far from the
event and to the GNSS network not particularly
dense over the Tuscany region (see Figure 10), then
not able to steer the simulated event, triggered over
the sea area, towards the A6 area.
It is worth noting that the Italian Civil Protection
(and regionals civil protection departments) alert sys-
tem is based on a daily bulletin issued each day before
13 UTC reporting the forecasted situation for the
following 36 h for each Alert Area of each Region
(see Figure 9, panel A for the Tuscany region). In this
context, it is important to consider that the 18 UTC
GFS global model is available at 21.45 UTC and the
forecast runtime is 2h30, so that the forecast forced
by the 18 UTC GFS is available at 00:15/00:30 UTC.
Along the same lines, the forecast initialized with 00
UTC GFS is available at 6:30 UTC. In an operational
framework only these two initializations are relevant
for issuing the daily alert bulletin, while subsequent
initializations at 06 UTC and 12 UTC (available,
respectively, at about 12.30 and 20.30 UTC) can be
used only for an alert update in a nowcasting frame-
work. Then, to gain a further insight in the eect of
the ingestion of the EO-derived products, the lower
panels of Figure 10 include also the hourly mean
rainfall for a new CTRL run started at 00UTC on
9 September 2017, i.e., close to the event onset
(CTRL_00UTC hereafter).
It can be seen that even the CTRL_00UTC was not
able to fully capture the observed rainfall peaks and
their timing, over the target areas S3 and A6.
Moreover, the WIND+SM+ZTD ingestion setup fol-
lowed more closely the CTRL_00UTC simulation
than the CTRL one. Finally, and more importantly,
the WIND+SM+ZTD ingestion setup produced
higher rainfall peaks with respect to the
CTRL_00UTC simulation, especially over the S3
area. Hence, the WIND+SM+ZTD experiment
resulted into a win-win situation being able to pro-
vide a better QPF over the target Alert Areas even in
advance with respect to one that would have been
available by forcing WRF with a more recent GFS
model instance.
The assimilation of Sentinel-derived products into
numerical weather prediction models was rarely
Figure 10. GNSS and EPN (European Permanent Network) stations used in this work with zoom on the stations placed in
attempted so far and can be considered as a challenging
problem related to the improvement of the quality of the
predictions of high-resolution numerical weather mod-
els. In the framework of the STEAM project, a number of
experiments was accomplished, making reference to
adisastrousood occurred in Italy in 2017, to verify the
eect of the ingestion of surface soil moisture, wind speed
and direction, land and sea surface temperature, as well as
precipitable water vapour in the WRF model. The work
has been performed trying to answer to the ESA investi-
gation about the value added obtained from the synergic
use of Sentinel-derived observation and NWP models.
By analyzing a set of sensitivity experiments con-
sidering one variable at a time, the most positive
impact on the WRF forecasts was achieved by assim-
ilating the wind speed and direction included in the
Ocean Wind Field component of the Sentinel-1
Level-2 Ocean product. This nding is likely related
to the fact that the triggering of the considered event
is strictly due to the presence of a convergence line
over the sea area in front of Livorno city. Thus, the
assimilation of a high spatial resolution wind eld
observation over the sea area allowed the model
dynamic to evolve in a way more similar to the
observed convective event with respect to the CTRL
simulation. The assimilation of GNSS-derived zenith
total delay turned out to have a positive impact as
well, except for an increase of the false alarm rate. For
the latter parameter, a decrease was obtained by
ingesting the Sentinel-1-derived SM. The ingestion
of LST produced a slight improvement of the quality
of the WRF predictions, while SST did not show any
positive inuence on the forecasts.
It must be underlined that the results of the assim-
ilation of the GNSS-derived ZTD are promising even
for the assimilation of high-resolution water vapour
maps derived from atmospheric phase screen maps
generated by applying the interferometric technique
to Sentinel-1 data. This application was not carried
out in this study, because of the very low values of the
water vapour in the atmosphere at the time of the
Sentinel-1 observation of the area of interest.
A further improvement of the accuracy of the
forecasts was achieved through a combined ingestion
of W
and W
, ZTD and SM. LST was not considered
in this experiment, despite its positive impact on the
WRF simulations, because, when high impact
weather events occur, cloud cover often hampers the
retrieval of this variable.
The outcomes of this study seem to conrm the use-
fulness of the synergy between high-resolution numerical
weather models and a set of Earth Observation products
at least for the forecast of high impact weather events.
Nonetheless, further investigations are required and will
be accomplished in the framework of the STEAM project.
Firstly, more than one case study has to be considered,
although high-resolution numerical weather predictions
are very expensive from a computational point of view.
Secondly, an assessment of the value added assimilating
high-resolution water vapour maps derived from atmo-
spheric phase screen maps generated by applying the
interferometric technique to Sentinel-1 data will be inves-
tigated in dierent cases to see the impact of higher spatial
resolution ZTD observations.
Disclosure statement
No potential conict of interest was reported by the
This work was supported by the European Space Agency
Martina Lagasio
Attema, E.P.W., & Ulaby, F.T. (1978). Vegetation modeled
as a water cloud. Radio Science,13(2), 357364.
Balenzano, A., Mattia, F., Satalino, G., & Davidson, M.W.J.
(2011). Dense temporal series of C- and L-band SAR
data for soil moisture retrieval over agricultural crops.
IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing,4(2), 439450.
Balenzano, A., Satalino, G., Lovergine, F., Rinaldi, M.,
Iacobellis, V., Mastronardi, N., & Mattia, F. (2013). On
the use of temporal series of L- and X-band SAR data for
soil moisture retrieval. Capitanata plain case study.
European Journal of Remote Sensing,46(1), 721737.
Beljaars, A.C.M. (1995). The parametrization of surface
uxes in large-scale models under free convection.
Quarterly Journal of the Royal Meteorological Society.
Bevis, M., Businger, S., Chiswell, S., Herring, T.A.,
Anthes, R.A., Rocken, C., & Ware, R.H. (1994). GPS
Meteorology: Mapping zenith wet delays onto precipita-
ble water. Journal of Applied Meteorology. doi:10.1175/
Bindlish, R., & Barros, A.P. (2001). Parameterization of
vegetation backscatter in radar-based, soil moisture
estimation. Remote Sensing of Environment,76(1),
130137. doi:10.1016/S0034-4257(00)00200-5
Böhm, J., & Schuh, H. (2004). Vienna mapping functions
in vlbi analyses. Geophysical Research Letters,31(1).
Böhm, J., Niell, A., Tregoning, P., & Schuh, H. (2006).
Global mapping function (gmf): a new empirical map-
ping function based on numerical weather model data.
Geophysical Research Letters,33(7).
Bouttier, F., & Courtier, P. (1999). Data assimilation con-
cepts and methods March 1999. Meteorological Training
Course Lecture Series. doi:10.1029/2007GL030733
Cardoso, R.M., Soares, P.M.M., Miranda, P.M.A., & Belo-
Pereira, M. (2013). WRF high resolution simulation of
Iberian mean and extreme precipitation climate.
International Journal of Climatology,33(11),
25912608. doi:10.1002/joc.3616
Dash, S.K., Sahu, D.K., & Sahu, S.C. (2013). Impact of AWS
observations in WRF-3DVAR data assimilation system:
A case study on abnormal warming condition in Odisha.
Natural Hazards. doi:10.1007/s11069-012-0393-0
Davis, C.A., Brown, B.G., Bullock, R., & Halley-Gotway, J.
(2009). The Method for Object-Based Diagnostic
Evaluation (MODE) applied to numerical forecasts
from the 2005 NSSL/SPC Spring Program. Weather
and Forecasting,24(5), 12521267. doi:10.1175/
Dyer, A.J., & Hicks, B.B. (1970). Flux-gradient relationships
in the constant ux layer. Quarterly Journal of the Royal
Meteorological Society. doi:10.1002/qj.49709641012
El Hajj, M., Baghdadi, N., Zribi, M., Belaud, G.,
Cheviron, B., Courault, D., & Charron, F. (2016).
Remote sensing of environment soil moisture retrieval
over irrigated grassland using X-band SAR data. Remote
Sensing of Environment,176, 202218. doi:10.1016/j.
Fiori, E., Comellas, A., Molini, L., Rebora, N., Siccardi, F.,
Gochis, D.J., . . . Parodi, A. (2014). Analysis and hindcast
simulations of an extreme rainfall event in the
Mediterranean area: The Genoa 2011 case. Atmospheric
Research. doi:10.1016/j.atmosres.2013.10.007
Fiori, E., Ferraris, L., Molini, L., Siccardi, F.,
Kranzlmueller, D., & Parodi, A. (2017). Triggering and
evolution of a deep convective system in the
Mediterranean Sea: Modelling and observations at
a very ne scale. Quarterly Journal of the Royal
Meteorological Society. doi:10.1002/qj.2977
Hajnsek, I., Jagdhuber, T., Member, S., Schön, H.,
Papathanassiou, K.P., & Member, S. (2009). Potential
of estimating soil moisture under vegetation cover by
means of PolSAR. IEEE Transactions on Geoscience and
Remote Sensing,47(2), 442454. doi:10.1109/
Han, J., & Pan, H.-L. (2011). Revision of convection and vertical
diusion schemes in the NCEP global forecast system.
Weather and Forecasting.doi:10.1175/waf-d-10-05038
Heroux, P., Kouba, J., Collins, P., & Lahaye, F. (2001). Gps
carrier-phase point positioning with precise orbit pro-
ducts. In Proceedings of the KIS (pp. 5-8).
Hong, S.-Y., & Lim, J.-O.J. (2006). The WRF
single-moment 6-class microphysics scheme (WSM6).
Journal of the Korean Meteorological Society,43(2),
Hong, S.-Y., Noh, Y., & Dudhia, J. (2006). A New Vertical
Diusion Package with an Explicit Treatment of
Entrainment Processes. Monthly Weather Review.
Hornacek, M., Wagner, W., Sabel, D., Truong, H.-L.,
Snoeij, P., Hahmann, T., . . . Doubkova, M. (2012).
Potential for high resolution systematic global surface
soil moisture retrieval via change detection using
Sentinel-1. IEEE Journal of Selected Topics in Applied
Earth Observations and Remote Sensing,5(4),
13031311. doi:10.1109/JSTARS.2012.2190136
Huete, A., Didan, K., van Leeuwen, W., Miura, T., &
Glenn, E. (2011). MODIS vegetation indices. In Remote
Sensing and Digital Image Processing,11, 579602.
Iacono, M.J., Delamere, J.S., Mlawer, E.J., Shephard, M.W.,
Clough, S.A., & Collins, W.D. (2008). Radiative forcing
by long-lived greenhouse gases: Calculations with the
AER radiative transfer models. Journal of Geophysical
Research Atmospheres. doi:10.1029/2008JD009944
Ide, K., Courtier, P., Ghil, M., & Lorenc, A. (1997). Unied
notation for data assimilation: Operational, sequential
and variational. Journal of the Meteorological Society of
Japan. doi:10.5194/npg-18-49-2011
Iriza, A., Dumitrache, R.C., Lupascu, A., & Stefan, S.
(2016). Studies regarding the quality of numerical
weather forecasts of the WRF model integrated at
high-resolutions for the Romanian territory. Atmosfera,
29(1), 1121. doi:10.20937/ATM.2016.29.01.02
Jackson, T.J., Chen, D., Cosh, M., Li, F., Anderson, M.,
Walthall, C., . . . Hunt, E.R. (2004). Vegetation water
content mapping using Landsat data derived normalized
dierence water index for corn and soybeans. Remote
Sensing of Environment,92(4), 475482. doi:10.1016/j.
Jackson, T.J., Le Vine, D.M., Hsu, A.Y., Oldak, A.,
Starks, P.J., Swift, C.T., . . . Haken, M. (1999). Soil moist-
ure mapping at regional scales using microwave radio-
metry: The Southern Great plains hydrology experiment.
Geoscience and Remote Sensing, IEEE Transactions On,
37(5), 21362151. doi:10.1109/36.789610
Kain, J.S., Weiss, S.J., Bright, D.R., Baldwin, M.E., Levit, J.J.,
Carbin, G.W., . . . Thomas, K.W. (2008). Some practical
considerations regarding horizontal resolution in the
rst generation of operational convection-allowing
NWP. Weather and Forecasting. doi:10.1175/
Kain, J.S., Weiss, S.J., Levit, J.J., Baldwin, M.E., & Bright, D.
R. (2006). Examination of convection-allowing cong-
urations of the WRF model for the prediction of severe
convective weather: The SPC/NSSL spring program
2004. Weather and Forecasting. doi:10.1175/waf906.1
Kim, S.B., Tsang, L., Johnson, J.T., Huang, S., Van Zyl, J.J.,
& Njoku, E.G. (2012). Soil moisture retrieval using
time-series radar observations over bare surfaces. IEEE
Transactions on Geoscience and Remote Sensing,50(5
PART 2), 18531863. doi:10.1109/TGRS.2011.2169454
Koizumi, K., Ishikawa, Y., & Tsuyuki, T. (2005).
Assimilation of precipitation data to the JMA mesoscale
model with a four-dimensional variational method and
its impact on precipitation forecasts. SOLA. doi:10.2151/
Kouba, J. (2008). Implementation and testing of the
gridded vienna mapping function 1 (vmf1). Journal Of
Geodesy,82(4-5), 193-205.
Lagasio, M., Parodi, A., Procopio, R., Rachidi, F., &
Fiori, E. (2017). Lightning potential index performances
in multimicrophysical cloud-resolving simulations of a
back-building mesoscale convective system: The Genoa
2014 event. Journal of Geophysical Research. doi:10.1002/
Marini, J. W. (1972). Correction of satellite tracking data
for an arbitrary tropospheric prole. Radio Science,7
Mateus, P., Catalao, J., & Nico, G. (2017). Sentinel-1
Interferometric SAR mapping of precipitable water
vapor over a country-spanning area. IEEE Transactions
on Geoscience and Remote Sensing,55(5), 29932999.
Mateus, P., Miranda, P.M.A., Nico, G., Catalão, J., Pinto, P.,
& Tomé, R. (2018). Assimilating InSAR maps of water
vapor to improve heavy rainfall forecasts: A case study
with two successive storms. Journal of Geophysical
Research: Atmospheres,123(7), 33413355. doi:10.1002/
Mateus, P., Tomé, R., Nico, G., Member, S., & Catalão, J.
(2016). Three-dimensional variational assimilation of
InSAR PWV using the WRFDA model. IEEE
Transactions on Geoscience and Remote Sensing, 54(12),
7323-7330. doi:10.1109/TGRS.2016.2599219
Niell, A. E. (1996). Global mapping functions for the atmo-
sphere delay at radio wavelengths. Journal of Geophysical
Research:Solid Earth,101(B2), 3227-3246.
Oh, Y. (2004). Quantitative retrieval of soil moisture con-
tent and surface roughness from multipolarized radar
observations of bare soil surfaces. IEEE Transactions on
Geoscience and Remote Sensing,42(3), 596601.
Panegrossi, G., Ferretti, R., Pulvirenti, L., & Pierdicca, N.
(2011). Impact of ASAR soil moisture data on the MM5
precipitation forecast for the Tanaro ood event of April
2009. Natural Hazards and Earth System Science,11(12),
31353149. doi:10.5194/nhess-11-3135-2011
Paulson, C.A. (1970). The mathematical representation of
wind speed and temperature proles in the unstable
atmospheric surface layer. Journal of Applied
Meteorology. doi:10.1175/1520-0450(1970)009<0857:T-
Pichelli, E., Ferretti, R., Cimini, D., Panegrossi, G.,
Perissin, D., Pierdicca, N., . . . Rommen, B. (2015).
InSAR water vapor data assimilation into mesoscale
model MM5: Technique and pilot study. IEEE Journal
of Selected Topics in Applied Earth Observations and
Remote Sensing,8(8), 38593875. doi:10.1109/JSTARS.2-
Pierdicca, N., Pulvirenti, L., & Bignami, C. (2010). Soil
moisture estimation over vegetated terrains using multi-
temporal remote sensing data. Remote Sensing of
Environment,114(2), 440448. doi:10.1016/j.
Pierdicca, N., Pulvirenti, L., Bignami, C., & Ticconi, F.
(2013). Monitoring soil moisture in an agricultural test
site using SAR data: Design and test of a pre-operational
procedure. IEEE Journal of Selected Topics in Applied
Earth Observations and Remote Sensing,6(3),
11991210. doi:10.1109/JSTARS.2012.2237162
Pierdicca, N., Pulvirenti, L., & Pace, G. (2014). A prototype
software package to retrieve soil moisture from
sentinel-1 data by using a bayesian multitemporal
algorithm. IEEE Journal of Selected Topics in Applied
Earth Observations and Remote Sensing,7(1), 153166.
Realini, E., & Reguzzoni, M. (2013). Gogps: open source
software for enhancing the accuracy of low-cost recei-
vers by single-frequency relative kinematic positioning.
Measurement Science and Technology,24(11), 115010.
Schepanski, K., Knippertz, P., Fiedler, S., Timouk, F., &
Demarty, J. (2015). The sensitivity of nocturnal low-level
jets and near-surface winds over the Sahel to model
resolution, initial conditions and boundary-layer
set-up. Quarterly Journal of the Royal Meteorological
Society,141(689), 14421456. doi:10.1002/qj.2453
Skamarock, W.C., Klemp, J.B., Dudhi, J., Gill, D.O.,
Barker, D.M., Duda, M.G., . . . Powers, J.G. (2008).
A description of the advanced research WRF Version
3. Technical Report, (June), 113. doi:10.5065/D6DZ069T
Sugimoto, S., Crook, N.A., Sun, J., Xiao, Q., & Barker, D.M.
(2009). An examination of WRF 3DVAR radar data
assimilation on its capability in retrieving unobserved
variables and forecasting precipitation through obser-
ving system simulation experiments. Monthly Weather
Review. doi:10.1175/2009mwr2839.1
Wang, H., Huang, X.Y., Sun, J., Xu, D., Zhang, M., Fan, S.,
& Zhong, J. (2014). Inhomogeneous background error
modeling for WRF-var using the NMC method. Journal
of Applied Meteorology and Climatology. doi:10.1175/
Webb, E.K. (1970). Prole relationships: The log-linear
range, and extension to strong stability. Quarterly
Journal of the Royal Meteorological Society. doi:10.1002/
... For the last two decades, applications of IWV estimations derived from GNSS tropospheric path delays have focused on several main directions such as: validations of IWV measurements by radiosondes or remote sensing satellite sensor sets, such as GOME-2, MODIS, OMI, SEVIRI and AIRS, combined with IWV estimations by NWP models [21,22]; climate change studies and weather forecasting [23]; assimilation of GNSS-IWV estimations in NWP models, such as WRF model [24][25][26][27]. These studies used point measurements data assimilation, from radiosondes measurements or GPS ZWD estimations, leading to small improvements (5-10%), for example [24,27] in WRF forecasts. ...
... For the last two decades, applications of IWV estimations derived from GNSS tropospheric path delays have focused on several main directions such as: validations of IWV measurements by radiosondes or remote sensing satellite sensor sets, such as GOME-2, MODIS, OMI, SEVIRI and AIRS, combined with IWV estimations by NWP models [21,22]; climate change studies and weather forecasting [23]; assimilation of GNSS-IWV estimations in NWP models, such as WRF model [24][25][26][27]. These studies used point measurements data assimilation, from radiosondes measurements or GPS ZWD estimations, leading to small improvements (5-10%), for example [24,27] in WRF forecasts. ...
Full-text available
Improving the accuracy of numerical weather predictions remains a challenging task. The absence of sufficiently detailed temporal and spatial real-time in-situ measurements poses a critical gap regarding the proper representation of atmospheric moisture fields, such as water vapor distribution, which are highly imperative for improving weather predictions accuracy. The estimated amount of the total vertically integrated water vapor (IWV), which can be derived from the attenuation of global positioning systems (GPS) signals, can support various atmospheric models at global, regional, and local scales. Currently, several existing atmospheric numerical models can estimate the IWV amount. However, they do not provide accurate results compared with in-situ measurements such as radiosondes. Here, we present a new strategy for assimilating 2D IWV regional maps estimations, derived from combined GPS and METEOSAT satellite imagery data, to improve Weather Research and Forecast (WRF) model predictions accuracy in Israel and surrounding areas. As opposed to previous studies, which used point measurements of IWV in the assimilation procedure, in the current study, we assimilate quasi-continuous 2D GPS IWV maps, combined with METEOSAT-11 data. Using the suggested methodology, our results indicate an improvement of more than 30% in the root mean square error (RMSE) of WRF forecasts after assimilation relative standalone WRF, when both are compared to the radiosonde measured data near the Mediterranean coast. Moreover, significant improvements along the Jordan Rift Valley and Dead Sea Valley areas are obtained when compared to 2D IWV regional maps estimations. Improvements in these areas suggest the impact of the assimilated high resolution IWV maps, with initialization times which coincide with the Mediterranean Sea Breeze propagation from the coastline to highland stations, as the distance to the Mediterranean Sea shore, along with other features, dictates its arrival times.
... It features multiple dynamical cores, a three-dimensional variational data assimilation system (3DVAR), and a software architecture allowing for computational parallelism and system extensibility. WRF is suitable for a broad spectrum of applications across scales ranging from metres to thousands of kilometres, and has been used by the authors for hydro-meteorological research applications in tropical and subtropical areas (Parodi and Tanelli 2010;Viterbo et al. 2016;Marras et al. 2017), and for mid-latitude severe weather studies (Parodi et al. 2012;Fiori et al. 2014;Viterbo et al. 2016;Fiori et al. 2017;Lagasio et al. 2017;Lagasio et al. 2019a;Lagasio et al. 2019b;Lagasio et al. 2019c;Parodi et al. 2019;Silvestro et al. 2019;Meroni et al. 2018b). ...
... The setup used for this work has been tested and used for operational applications by CIMA Foundation for ICPD and for ARPAL, as well as in previous research works (Lagasio et al. 2019a;Lagasio et al. 2019b;Lagasio et al. 2019c): it consists of a configuration with three two-way nested domains with horizontal grid spacing of 13.5, 4.5, and 1.5 km and with 50 vertical levels (Fig. 3). ...
Full-text available
Between the 4 th and the 6 th of November 1994, Piedmont and the western part of Liguria (two regions in northwestern Italy) were hit by heavy rainfalls that caused the flooding of the Po, the Tanaro rivers and several of their tributaries, causing 70 victims and the displacement of over 2000 people. At the time of the event, no early warning system was in place and the concept of hydro-meteorological forecasting chain was in its infancy, since it was still limited to a reduced number of research applications, strongly constrained by coarse-resolution modelling capabilities both on the meteorological and the hydrological sides. In this study, the skills of the high-resolution CIMA Research Foundation operational hydro-meteorological forecasting chain are tested in the Piedmont 1994 event. The chain includes a cloud-resolving numerical weather prediction (NWP) model, a stochastic rainfall down-scaling model, and a continuous distributed hydrological model. This hydro-meteorological chain is tested in a set of operational configurations, meaning that forecast products are used to initialise and force the atmospheric model at the boundaries. The set consists of four experiments with different options of the microphysical scheme, which is known to be a critical parameterisation in this kind of phenomena. Results show that all the configurations produce an adequate and timely forecast (about 2 days ahead) with realistic rainfall fields and, consequently, very good peak flow discharge curves. The added value of the high resolution of the NWP model emerges, in particular, when looking at the location of the convective part of the event, which hit the Liguria region.
... Вопросам повышения эффективности численных моделей прогноза погоды за счет ассимиляции в них данных ДЗЗ посвящено большое количество публикаций зарубежных авторов [1][2][3][4][5][6][7]. При этом, далеко не во всех работах достигается однозначно положительный эффект в части оправдываемости прогнозов метеорологических моделей, а результаты этих работ имеют строгую региональную привязку и не могут быть обобщены или перенесены на другие регионы. ...
The problem of improving the WRF numerical weather model performance for the territory of Belarus by assimilating the Earth remote sensing data is considered. It is shown that for the winter period, the use of satellite data of high spatial resolution, including on the structure of land use , albedo, leaf index and photosynthetically active radiation absorbed by the underlying surface can reduce a root-mean-square error of the short-term forecast (up to 48 h) of the air surface temperature by 0.53–1.11 °С. For the summer period, on the basis of numerical experiments the optimal correction factor for the land surface albedo was estimated. This made it possible to reduce a root-mean-square error of temperature forecast at the meteorological stations of Belarus for the lead time of +12, +24, +36, and +48 h by an average of 0.30 °С, 0.10 °С, 0.15 °С, and 0.16 °С, respectively.
... Moreover, experimental results obtained by feeding NWP models and black box models with ZTD values retrieved by GNSS will be shown. Some of the presented results and methodologies are published in the co-authored articles [95], [94], [57] and [56]. ...
... has been exploited, using cutting edge observations and techniques. For example, a new operator for the assimilation of radar data was developed by Lagasio et al. (2019c), or high resolution satellite-derived maps were assimilated by Lagasio et al. (2019b) and Lagasio et al. (2019a), showing a positive impact on the forecast of heavy rainfall events in Italy. CIMA is also actively involved in research activities about the impacts that the assimilation of high resolution water vapor maps has on the forecast of heavy rains . ...
Full-text available
Numerical models are operationally used for weather forecasting activities to reduce the risks of several hydro-meteorological disasters. The overarching goal of this work is to evaluate the Weather Research and Forecasting (WRF) model predictive capabilities over the Italian national territory in the year 2018, in two specific cloud resolving configurations. The validation is carried out with a fuzzy logic approach, by comparing the precipitation predicted by the WRF model, and the precipitation observed by the national network. The fuzzy logic technique, by considering different intensity thresholds, allows to identify the reliable spatial scales of the forecasts. The same approach is applied to evaluate the performances of COSMO-2I model, a state-of-the-art numerical model configuration used for operational activities. For the entire year, except for summer, the model predictive capabilities are high, with useful forecasts for structures of medium intensities down to O (10 km) length scales. In summer the skills decrease mainly because of localization errors. The work aims to provide a robust evaluation of the forecast performances of another convection permitting operational meteorological models currently available in Italy.
Full-text available
Soil moisture (SM) datasets at high spatial resolutions are beneficial for a wide range of applications, such as monitoring and prediction of hydrological extremes, numerical weather prediction, and precision agriculture. For large scale applications in particular, remotely sensed SM has advantages over in situ data because it provides gridded estimates and because it is less labour-intensive. However, until present, active microwave SM data have not been presented at their native spatial resolution, since the quality of these data is limited by speckle. We explored the potential and limits of high spatial resolution of active microwave SM observations. We used a Sentinel-1 C-band SAR SM dataset at six spatial resolutions ranging from 20 × 20 to 120 × 120 m. This was compared to a closely spaced (20 m) in situ dataset collected on a non-irrigated agricultural field (2.5 ha) in the Southeast of Luxembourg. A comparison of the field and satellite datasets demonstrated how Sentinel-1 data with a high spatial resolution can be used to quantify temporal within-field SM variability. SM was accurately estimated at spatial resolutions of 60 × 60 m and coarser, where the temporal correlation was found to be 0.67 and sub-field variations in SM were still detected. Spatial correlation was limited by the absence of SM variability within the field. These results indicate that high spatial resolution SM estimates from Sentinel-1 data can be valuable for monitoring temporal SM variations within agricultural fields.
Full-text available
Predicting extreme weather events in a short time period and their developing in localized areas is a challenge. The nowcasting of severe and extreme weather events is an issue for air traffic management and control because it affects aviation safety, and determines delays and diversions. This work is part of a larger study devoted to nowcasting rain and wind speed in the area of Malpensa airport by merging different datasets. We use as reference the weather station of Novara to develop a nowcasting machine learning model which could be reusable in other locations. In this location we have the availability of ground-based weather sensors, a Global Navigation Satellite System (GNSS) receiver, a C-band radar and lightning detectors. Our analysis shows that the Long Short-Term Memory Encoder Decoder (LSTM E/D) approach is well suited for the nowcasting of meteorological variables. The predictions are based on 4 different datasets configurations providing rain and wind speed nowcast for 1 h with a time step of 10 min. The results are very promising with the extreme wind speed probability of detection higher than 90%, the false alarms lower than 2%, and a good performance in extreme rain detection for the first 30 min. The configuration using just weather stations and GNSS data in input provides excellent performances and should be preferred to the other ones, since it refers to the pre-convective environment, and thus can be adaptable to any weather conditions.
Full-text available
The model initialization with high-resolution SAR wind data provided by the Sentinel-1 mission and its impact on the meteorological model WRF-ARW simulations is discussed. The activity is performed within the Horizon 2020 CEASELESS project, focusing on one of the target areas, the northern Adriatic Sea (northern-central Mediterranean). The Sentinel-1 SAR wind is ingested into LAPS, a numerical system developed at NOAA, specifically designed for data analysis and nowcasting issues, since it has the advantage of being faster and less computational demanding than advanced data assimilation methods. Here, LAPS analyses are used to perform a smarter initialization of the WRF-ARW model simulations than using simply global model fields. The impact of the Sentinel-1 SAR wind on the model simulations is evaluated for twenty cases, ranging through several atmospheric conditions occurring in different seasons of the years 2014–2018. For each case study, a reference WRF-ARW simulation is forced with GFS analysis and forecasts used as initial and boundary conditions, respectively. Additional model runs are initialized with the LAPS analyses, which include the information of Sentinel-1 SAR wind, METAR data and the SEVIRI/MSG (Eumetsat) brightness temperature. A statistical evaluation of the WRF-ARW simulations is performed versus an independent set of surface records, provided by the Friuli Venezia Giulia regional station network (northeastern Italy), and METAR data. The comparison is performed for 10 m wind, 2 m air and dew point temperature. The results show a positive, albeit modest, impact on the WRF model simulations initialized with the LAPS analyses. The initialization with the Sentinel-1 SAR wind show benefits for all surface variables. Finally, a Mediterranean tropical-like cyclone (Medicane), occurred in the Ionian Sea in November 2017, is considered in order to show how the use of Sentinel wind data can contribute to a better analysis and simulation of severe weather episodes in the Mediterranean. The improvement in the simulation of the pressure minimum location is remarkable.
Full-text available
Gaining a deeper physical understanding of the high‐impact weather events which repeatedly affected the Western Mediterranean Basin in recent years on the coastal areas of eastern Spain, s outhern France and northern Italy is strongly motivated by the social request to reduce the casualties and the economical impacts due to these highly localized and hardly predictable phenomena. In October 2014, an extreme event hit Genoa city centre, less than 3 years after a very similar event, which occurred in November 2011. Taking advantage of the availability of both observational data and modelling results at the micro‐α meteorological scale, this article provides insights about the triggering mechanism and the subsequent spatio‐temporal evolution of the Genoa 2014 back‐building Mesoscale Convective System. The major finding is the effect of a virtual mountain created over the Ligurian Sea by the convergence of a cold and dry jet outflowing from the Po valley and a warm and moist low‐level southeasterly jet within the planetary boundary layer .
Full-text available
The aim of this paper is to evaluate the quality of high-resolution weather forecasts from the Weather Research and Forecasting (WRF) numerical weather prediction model. The lateral and boundary conditions were obtained from the numerical output of the Consortium for Small-scale Modeling (COSMO) model at 7 kmhorizontal resolution. The WRF model was run for January and July 2013 at two horizontal resolutions(3 and 1 km). The numerical forecasts of the WRF model were evaluated using different statistical scores for 2 m temperature and 10 m wind speed. Results showed a tendency of the WRF model to overestimate the values of the analyzed parameters in comparison to observations.
Very high-resolution Precipitable Water Vapor (PWV) maps obtained by the Sentinel-1 A Synthetic Aperture Radar (SAR), using the SAR interferometry (InSAR) technique, are here shown to have a positive impact on the performance of severe weather forecasts. A case study of deep convection which affected the city of Adra, Spain, on 6-7 September 2015, is successfully forecasted by the Weather Research and Forecasting (WRF) model initialized with InSAR data assimilated by the three-dimensional variational (3D-Var) technique, with improved space and time distributions of precipitation, as observed by the local weather radar and rain gauge. This case study is exceptional because it consisted of two severe events 12 hours apart, with a timing that allows for the assimilation of both the ascending and descending satellite images, each for the initialization of each event. The same methodology applied to the network of Global Navigation Satellite System (GNSS) observations in Iberia, at the same times, failed to reproduce observed precipitation, although it also improved, in a more modest way, the forecast skill. The impact of PWV data is shown to result from a direct increment of convective available potential energy, associated with important adjustments in the low-level wind field, favoring its release in deep convection. It is suggested that InSAR images, complemented by dense GNSS data, may provide a new source of water vapor data for weather forecasting, since their sampling frequency could reach the sub-daily scale by merging different SAR platforms, or when future geosynchronous radar missions become operational.
Severe weather events are responsible for hundreds of fatalities and millions of euros of damage every year on the Mediterranean basin. Lightning activity is a characteristic phenomenon of severe weather and often accompanies torrential rainfall, which, under certain conditions like terrain type, slope, drainage, soil saturation may turn into flash flood. Building on the existing relationship between significant lightning activity and deep convection and precipitation, the performance of the Lightning Potential Index (LPI), as a measure of the potential for charge generation and separation that leads to lightning occurrence in clouds, is here evaluated for the V-shape back-building Mesoscale Convective System which hit Genoa city (Italy) in 2014. An ensemble of WRF simulations at cloud-permitting grid spacing (1 km) with different microphysics parameterizations is performed and compared to the available observational radar and lightning data. The results allow gaining a deeper understanding of the role of lightning phenomena in the predictability of V-shape back-building MCSs often producing flash flood over western Mediterranean complex topography areas. Moreover, they support the relevance of accurate lightning forecasting for the predictive ability of these severe events.
This paper presents a methodology to generate maps of atmosphere's precipitable water vapor (PWV) over large areas with a length of hundreds of kilometers and a width of about 250 km, based on the use of interferometric Sentinel-1A/B C-band synthetic aperture radar (SAR) data with a high spatial resolution of $5 \times 20$ m$^2$ and the revisiting time of six days. An algorithm to calibrate and merge PWV maps from different swaths of Sentinel-1 acquired along the same track, using global navigation satellite system (GNSS) measurements, is described. The proposed methodology is tested on Sentinel-1A SAR images acquired over the Iberian Peninsula, along both descending and ascending tracks. The assessment with an independent set of GNSS measurements shows a mean difference of a fraction of millimeter and a dispersion lower than 2 mm. Both the use of Sentinel-1A/B SAR images and the proposed methodology open new perspectives on the application of SAR meteorology for the high-resolution mapping of PWV over large region-spanning areas and the assimilation of interferometric SAR data into numerical weather models.
This paper studies the problem of the assimilation of precipitable water vapor (PWV), estimated by synthetic aperture radar interferometry, using the Weather Research and Forecast Data Assimilation model 3-D variational data assimilation system. The experiment is designed to assess the impact of the PWV assimilation on the hydrometers and the rainfall predictions during 12 h after the assimilation time. A methodology to obtain calibrated maps of PWV and estimated their precision is also presented. The forecasts are compared with GPS estimates of PWV and with rainfall observations from a meteorological radar. Results show that after data assimilation, there is a correction of the bias in the PWV prediction and an improvement in the prediction of the weak to moderate rainfall up to 9 h after the assimilation time.
The aim of this study was to develop an inversion approach to estimate surface soil moisture from X-band SAR data over irrigated grassland areas. This approach simulates a coupling scenario between Synthetic Aperture Radar (SAR) and optical images through the Water Cloud Model (WCM). A time series of SAR (TerraSAR-X and COSMO-SkyMed) and optical (SPOT 4/5 and LANDSAT 7/8) images were acquired over an irrigated grassland region in southeastern France. An inversion technique based on multi-layer perceptron neural networks (NNs) was used to invert the Water Cloud Model (WCM) for soil moisture estimation. Three inversion configurations based on SAR and optical images were defined: (1) HH polarization, (2) HV polarization, and (3) both HH and HV polarizations, all with one vegetation descriptor derived from optical data. The investigated vegetation descriptors were the Normalized Difference Vegetation Index "NDVI", Leaf Area Index "LAI", Fraction of Absorbed Photosynthetically Active Radiation "FAPAR", and the Fractional vegetation COVER "FCOVER". These vegetation descriptors were derived from optical images. For the three inversion configurations, the NNs were trained and validated using a noisy synthetic dataset generated by the WCM for a wide range of soil moisture and vegetation descriptor values. The trained NNs were then validated from a real dataset composed oTERRASAR-X COSMO-SKYMED X-band SAR backscattering coefficients and vegetation descriptor derived from optical images. The use of X-band SAR measurements in HH polarization (in addition to one vegetation descriptor derived from optical images) yields more precise results on soil moisture (Mv) estimates. In the case of NDVI derived from optical images as the vegetation descriptor, the Root Mean Square Error on Mv estimates was 3.6 Vol.% for NDVI values between 0.45 and 0.75, and 6.1 Vol.% for NDVI between 0.75 and 0.90. Similar results were obtained regardless of the other vegetation descriptor used.