Detection of water vapor time variations associated with heavy rain in northern Italy by geodetic and low-cost GNSS receivers

Abstract

GNSS atmospheric water vapor monitoring is not yet routinely performed in Italy, particularly at the regional scale. However, in order to support the activities of regional environmental protection agencies, there is a widespread need to improve forecasting of heavy rainfall events. Localized convective rain forecasts are often misplaced in space and/or time, causing inefficiencies in risk mitigation activities. Water vapor information can be used to improve these forecasts. In collaboration with the environmental protection agencies of the Lombardy and Piedmont regions in northern Italy, we have collected and processed GNSS and weather station datasets for two heavy rain events: one which was spatially widespread, and another which was limited to few square kilometers. The time variations in water vapor derived from a regional GNSS network with inter-station distances on the order of 50 km were analyzed, and the relationship between the time variations and the evolution of the rain events was evaluated. Results showed a signature associated with the passage of the widespread rain front over each GNSS station within the area of interest. There was a peak in the precipitable water vapor value when the heavier precipitation area surrounded the station, followed by a steep decrease (5–10 mm in about 1 h) as the rainclouds moved past the station. The smaller-scale event, a convective storm a few kilometers in extent, was not detected by the regional GNSS network, but strong fluctuations in water vapor were detected by a low-cost station located near the area of interest.

Full text

Barindelli et al. Earth, Planets and Space (2018) 70:28
https://doi.org/10.1186/s40623-018-0795-7
FULL PAPER
Detection of water vapor time variations
associated with heavy rain in northern Italy
by geodetic and low-cost GNSS receivers
Stefano Barindelli1, Eugenio Realini2* , Giovanna Venuti1, Alessandro Fermi1 and Andrea Gatti1
Abstract
GNSS atmospheric water vapor monitoring is not yet routinely performed in Italy, particularly at the regional scale.
However, in order to support the activities of regional environmental protection agencies, there is a widespread need
to improve forecasting of heavy rainfall events. Localized convective rain forecasts are often misplaced in space and/
or time, causing inefficiencies in risk mitigation activities. Water vapor information can be used to improve these fore-
casts. In collaboration with the environmental protection agencies of the Lombardy and Piedmont regions in north-
ern Italy, we have collected and processed GNSS and weather station datasets for two heavy rain events: one which
was spatially widespread, and another which was limited to few square kilometers. The time variations in water vapor
derived from a regional GNSS network with inter-station distances on the order of 50 km were analyzed, and the rela-
tionship between the time variations and the evolution of the rain events was evaluated. Results showed a signature
associated with the passage of the widespread rain front over each GNSS station within the area of interest. There was
a peak in the precipitable water vapor value when the heavier precipitation area surrounded the station, followed by
a steep decrease (5–10 mm in about 1 h) as the rainclouds moved past the station. The smaller-scale event, a con-
vective storm a few kilometers in extent, was not detected by the regional GNSS network, but strong fluctuations in
water vapor were detected by a low-cost station located near the area of interest.
Keywords: GNSS meteorology, PWV variations, Intense rainfall
© The Author(s) 2018. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License
(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium,
provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license,
and indicate if changes were made.
Open Access
*Correspondence: eugenio.realini@g-red.eu
2 Geomatics Research and Development (GReD) srl, via Cavour 2,
22074 Lomazzo, Como, Italy
Full list of author information is available at the end of the article
Introduction
Global navigation satellite systems (GNSS) meteorology
(Bevis et al. 1992) relies on estimating the GNSS signal
propagation delays induced by the presence of water
vapor in the lower troposphere. GNSS-based tropo-
spheric water vapor monitoring is used for operational
meteorology in only a few countries, e.g., Japan (JMA
2013), the UK, and France (Bennitt and Jupp 2012; Guer-
ova etal. 2016). In these countries, time series of zenith
total delays (ZTD) or precipitable water vapor (PWV),
derived from GNSS observations which are continu-
ously collected by permanent regional networks of geo-
detic receivers, are routinely assimilated into mesoscale
numerical weather prediction (NWP) models (Bennitt
and Jupp 2012; De Haan 2013; Saito etal. 2017). e use
of GNSS for locally intense rainfall prediction is a current
topic of research. In particular, the ability of hyper-dense
GNSS networks (with receivers only a few kilometers
apart) to provide observations at the necessary high spa-
tial resolution is under investigation (Shoji et al. 2004;
Deng etal. 2009; Sato etal. 2013). e hyper-dense net-
works can be supplemented by integrating GNSS-derived
information with atmospheric artifacts retrieved from
synthetic aperture radar (SAR) images (Onn 2006). us
far, this solution has been realized only at an experimen-
tal level, but hyper-dense networks, possibly comple-
mented by SAR-derived water vapor maps, are expected
to soon be widespread at an operational level thanks to
the augmented potentialities of GNSS system, the grow-
ing number of permanent stations, and the availability of
increasingly high-performing low-cost receivers. Some

Page 2 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
of the authors of this paper are involved in this research
and are working both on the deployment of dense net-
works of single-frequency receivers, which are already
available on the market (within the H2020 project BRIG-
AID—http://brigaid.eu/), and on the development of
low-cost dual-frequency sensors (EUROSTARS project
EDWIGE—https://www.eurostars-eureka.eu/project/
id/10235).
Although Italy is endowed with numerous perma-
nent GNSS stations for geodetic and surveying pur-
poses, GNSS meteorology has not yet seen use in routine
weather prediction and is still solely the domain of
research. at research has taken several forms. Long-
term ZTD time series derived from the European per-
manent network (EPN) have been analyzed (Pacione
etal. 2013), and experimental studies on the use of GNSS
data for the tomographic reconstruction of the refractiv-
ity field have been conducted (Notarpietro et al. 2011).
GNSS-derived water vapor has been compared to other
water vapor products (Basili etal. 2001; Boccolari etal.
2002; Ciotti etal. 2003; Cheng etal. 2012; Bonafoni and
Biondi 2016). Finally, within the last decade, a few stud-
ies on the use of GNSS products for both regional- and
local-scale meteorological events have been carried out
(Faccani etal. 2005; Tammaro etal. 2016;Ferrando 2017).
Within this framework, we report on an experimental
activity in northern Italy aimed at promoting the use of
GNSS meteorology on an operational level. GNSS ZTD
estimates and weather data were used to derive time
series of PWV during a period in which two local, intense
rainfall events occurred. e goal was to detect PWV
time variations and evaluate their relationship to those
events. Furthermore, an innovative potential solution to
the need for higher spatial resolution water vapor moni-
toring was investigated: using low-cost GNSS sensors to
densify existing networks.
e Piedmont and Lombardy regions in northwest-
ern Italy are endowed with a GNSS permanent network,
which is managed by an inter-regional positioning service
called SPIN. e service freely delivers GNSS raw dual-
frequency data at a 30-s rate. Two networks of weather
stations collecting surface temperature and pressure data
are present in the same area and are managed by two dif-
ferent regional environmental protection agencies (ARPA
Lombardia and ARPA Piemonte). e two weather net-
works are integrated neither with one another nor with
the GNSS network. Moreover, no Italian GNSS service
delivers tropospheric delay or water vapor products,
either for research or for operational purposes.
e GNSS data collected from the SPIN network dur-
ing the two rain events were used to estimate ZTD time
series by means of a free and open source software for
GNSS data processing. e estimates were validated
against results from state-of-the-art GNSS software.
By using the estimated ZTD and properly interpolated
weather data, PWV time series were computed and vali-
dated against independent radiosonde-derived PWV val-
ues. A first analysis of their variations in connection with
the rain events was then carried out. Finally, the perfor-
mance of a low-cost single-frequency GNSS station was
investigated by a preliminary comparison of its PWV
values against those of a near dual-frequency geodetic
station.
In the following sections, we start by introducing the
experimental setup and the processing used to obtain
ZTD and GNSS-derived PWV time series from the
GNSS and meteorological data. e internal and exter-
nal validation procedures follow, and the analysis of the
PWV variations and final considerations conclude the
paper.
Experimental setup and data processing
e SPIN GNSS network comprises 30 permanent sta-
tions with inter-station distances of about 50km, uni-
formly distributed in the territories of the Piedmont and
Lombardy regions (Fig.1). During the period of interest,
we collected 30-s RINEX observation files from all SPIN
stations. In addition, we collected the RINEX files from
CATU station, which belongs to the nationwide net-
work NetGeo (deployed and managed by Topcon). e
CATU station data were used to gain further insight into
the local variations in water vapor in an area that was
affected by very localized and strong convective precipi-
tation, but where no SPIN receivers are present (see next
section). In the same area, a low-cost single-frequency
station (marker name GRED), deployed and continu-
ously operated by Geomatics Research & Development
srl, is present. e data collected from this station were
processed as well. e estimated ZTD were compared to
those obtained from the nearest CATU geodetic receiver
to evaluate to what extent a low-cost GNSS receiver and
antenna could be used to detect water vapor variations.
All geodetic stations are equipped with either Leica or
Topcon receivers and antennas. e low-cost GRED sta-
tion uses a u-blox LEA-6T receiver with a Tallysman L1
antenna.
e atmospheric pressure and temperature measure-
ments needed to convert the GNSS-estimated tropo-
spheric delay into PWV were obtained from two regional
networks of weather stations operated by ARPA Piemonte
and ARPA Lombardia environmental protection agen-
cies. Data provided by ARPA Lombardia have a temporal
resolution of 10min, while those provided by ARPA Pie-
monte have a temporal resolution of 30min. A subset of
35 stations was selected (Fig.2) according to a proximity
criterion with respect to the 30 GNSS stations. Further

Page 3 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
details about the GNSS and weather stations hardware
setup are reported in “Appendix.”
Figure3 shows the position of the two radiosonde sta-
tions used for validation of GNSS-derived PWV. One is
located in Lombardy [World Meteorological Organiza-
tion (WMO) station identifier 16080, Milano Linate] and
one in Piedmont (WMO station identifier 16113, Cuneo
Levaldigi). Both stations launch one radiosonde every
12h.
e GNSS observation processing was carried out by
means of the goGPS MATLAB open source software
(Realini and Reguzzoni 2013; Herrera etal. 2016), modi-
fied by the authors to perform precise point position-
ing (PPP) (Kouba and Héroux 2001) and estimate ZTD
values. e most recent version of the software, includ-
ing the PPP module and ZTD estimation capability, is
freely downloadable from the goGPS project GitHub
page (https://github.com/goGPS-Project). Since the
tropospheric delay estimation capability is a recent addi-
tion to the goGPS software, the final ZTD estimates
were validated not only against radiosonde values, but
also against an independent run of the state-of-the-art
Bernese software (Dach etal. 2015), which uses its own
PPP algorithm. goGPS and Bernese apply their PPP algo-
rithms in two different ways: Bernese implements a least-
squares post-processing adjustment of all observations
collected during a given time period (in this case, daily)
to jointly estimate coordinates and ZTD parameters, and
goGPS adopts an extended Kalman filter. It should be
noted that Caldera etal. (2016) already validated goGPS
against Bernese with regard to precise relative position-
ing (Teunissen and Kleusberg 1998; Hoffman-Wellen-
hof etal. 2001) and demonstrated that the two software
packages reach the same millimeter level of accuracy.
goGPS and Bernese were run using the same processing
parameters where possible: GPS-only observations, ion-
osphere-free combination of L1 and L2 observations,
10◦
for the satellites elevation cutoff, global mapping function
(Böhm etal. 2006), tropospheric gradient not estimated,
I08.ATX PCO/PCV model, solid Earth tides model
according to IERS convention, and ocean loading model
FES2004. goGPS was set to a 30-s processing rate, while
the Bernese processing rate was set to 10min, following
the recommendations regarding its PPP implementation.
Within goGPS, the coordinates dynamic model expresses
the fact that the stations are fixed, and both the receiver
Fig. 1 SPIN GNSS network

Page 4 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
clock and ZTD parameters are modeled as random walk
processes (Kouba and Héroux 2001). e sigma of the
former is set to a large value to account for its unpredict-
able behavior, while that of the latter is modeled as:
where
σh=1 mm/√h
. In the Bernese software, a new
ZTD parameter is introduced every 10min by linearly
modeling its evolution within this interval. erefore,
each parameter accounts for the contribution of the
tropospheric water vapor variations in the interval con-
sidered. Moreover, additional pseudo-observations are
introduced, expressing a regularity of the ZTD behavior:
the inclination of the linear function must be 0 with an a
priori given standard deviation. In the present case, this
standard deviation was set equal to 3mm. For the coor-
dinates, a unique set of parameters is introduced for the
whole daily time period under consideration.
Five of the 30 SPIN stations, namely ALSN, DEMN,
GOZZ, SERR, and VIGE, were excluded due to high
numbers of outliers and large gaps in code and phase
observations. e remaining 25 GNSS stations were
(1)
σ(ZTD
t
+�
t
−ZTD
t
)=√�
t
·σh
deemed sufficient both in number and in spatial distribu-
tion for the selected case studies.
Within the goGPS PPP algorithm, the ZTD update with
respect to the value of the previous epoch depends on all
the observations collected from the satellites which are
simultaneously in view. e ZTD models the common
projection of the delays affecting the considered GNSS
signals along the zenith direction above the receiver and
along their different receiver-satellite paths. e use of
observations confined to a restricted volume around the
zenith direction should in principle allow for a ZTD esti-
mate closer to the actual zenith value. Such restriction,
where possible, would weaken the acquisition system
geometry, degrading all parameter estimates. However,
all current GNSS software implements a weighted adjust-
ment that makes the solution less dependent on the
observations at low elevations above the horizon. In fact,
although the weighted adjustment improves the estimates
of the unknown parameters, such observations are nois-
ier, mainly due to multipath effects (Dach etal. 2015). A
weighted adjustment that expresses the observation accu-
racy as an increasing function of the elevation could be
exploited to improve detection of the actual tropospheric
Fig. 2 Weather stations network

Page 5 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
water vapor along the zenith direction above the receiver,
which is our aim. A better solution, albeit one requiring
a nonstandard analysis of the GNSS adjustment residu-
als, would be to reconstruct the satellite-receiver slant
delays and, after selecting only those above a given eleva-
tion threshold, average their zenith projection [as in Sato
etal. (2013)]. is approach presents some drawbacks: it
requires the development of ad hoc software; the recon-
structed slant delays could be affected by unmodeled
multipath effects; and finally, the averaged zenith pro-
jection, which relies on a limited number of slant delays,
is highly dependent on the sporadic contribution of a
satellite which is in view for only a short period. In this
paper we have thus decided to investigate the effect of
observation weighting on goGPS-derived ZTD estimates.
We expected the weighted solution to be more sensitive
to the local state of the troposphere. Two different runs
were performed: one without observation weighting and
one with satellite elevation-dependent weighting based
on the sine squared of the elevation angle.
To enable tropospheric delay estimation for the low-
cost L1-only receiver, the ionospheric delay for the GRED
station was interpolated from the three surrounding
dual-frequency stations (namely COMO, NOVR, and
MILA) by applying the Satellite-specific Epoch-differ-
enced Ionospheric Delay (SEID) algorithm (Deng et al.
2009).
While data provided by GNSS stations have a common
30-s observation rate, the temperature and pressure data
of the 35 available weather stations have different time
resolutions (30min for the Piemonte stations and 10min
for the Lombardia stations). ese data were thus inter-
polated to match the GNSS observation rate and to fill
some data gaps. is interpolation was done by means
of a modified least-squares collocation (LSC) algorithm.
e algorithm was applied to the time series of pressure
and temperature of each meteorological station, which
were modeled as random processes stationary within a
time interval that depends on the gap size. e required
covariance model was estimated from the data surround-
ing the prediction point.
e interpolated temperature and pressure time series
of the weather stations were then associated with the
GNSS stations according to the nearest neighbor cri-
terion, and height corrections were applied following
Realini etal. (2014). By exploiting the 30-s pressure time
Fig. 3 Radiosonde stations in the analyzed area. WMO id 16113: Cuneo Levaldigi station; WMO id 16080: Milano Linate station

Page 6 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Fig. 4 Time progression of the July 22, 2016, event (a, c, e) and the July 26, 2016, event (b, d, f), as seen through precipitation intensity maps
derived from radar reflectivity observations. Such maps are published by the Centro Meteo Lombardo-CML, based on MeteoSwiss radar data. LT
local time. a July 22—07:00 LT. b July 26—12:00 LT. c July 22—10:00 LT. d July 26—15:00 LT. e July 22—13:00 LT. f July 26—18:00 LT

Page 7 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
series, zenith hydrostatic delays (ZHD) were computed
(Saastamoinen 1973) and subtracted from the estimated
ZTD to obtain ZWD time series. e ZWD time series
were then transformed into the final PWV 30-s time
series following Askne and Nordius (1987). PWV val-
ues from ZTD are named
PWVgoGPS
and
PWVgoGPSw
if derived from equally weighted or elevation weighted
observation adjustments, respectively.
Case studies
Two different intense rain events that occurred on July 22
and July 26, 2016, over northern Italy were considered.
A qualitative description of the rain events is provided
by maps of rain intensity derived from radar reflectivity
observations (Fig.4). Figure5 shows the corresponding
maps of observed temperature, interpolated from the
weather station data by kriging. e July 22, 2016, event
Fig. 5 Time progression of the July 22, 2016, event (a, c, e) and the July 26, 2016, event (b, d, f), as seen through temperature maps derived from
weather stations observations. Temperature values in degrees Celsius. LT local time. a July 22—07:00 LT. b July 26—12:00 LT. c July 22—10:00 LT. d
July 26—15:00 LT. e July 22—13:00 LT. f July 26—18:00 LT

Page 8 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
was characterized by a widespread intense rain front that
moved from southwest to northeast, passing through the
Piedmont and Lombardy regions over about 8 h, from
06:00 to 14:00 local time (LT). In contrast, the July 26,
2016, rain event was characterized by sparse precipita-
tion cells that developed over these regions in a continu-
ous process of generation and dissipation that also lasted
for about 8h, from 12:00 to 20:00 LT. For this second
event, attention was focused on the convective cell reach-
ing cloudburst level at around 18:00 LT with both heavy
rain and hail. is cell is visible at the center of the radar
image in Fig.4f, represented with a color corresponding
to 110–440mm/h range.
Results and discussion
Validation of goGPS PWV against radiosonde and Bernese
PWV
Italian radiosonde stations are managed by the Italian
Air Force, which makes the observations available to the
European Centre for Medium-Range Weather Forecasts
(ECMWF). e University of Wyoming Web site provides
a convenient way to access and download global radio-
sonde data (Oolman 2017). For each radiosonde launch,
the integral PWV values are computed and are published
together with the sensed values along the vertical profile
above the radiosonde station. PWV values from the radi-
osonde stations of Milano Linate and Cuneo Levaldigi
were compared with PWV values derived from the two
GNSS stations of MILA and SAVI, respectively, from
July 16 to July 31, 2016. Two radiosonde launches per day
over 16 days resulted in 32 PWV data points. e height
difference between the GNSS and radiosonde stations is
limited to 10m for the first case study and is 53m for the
second case study. ese differences were accounted for
by using the Saastamoinen model for wet delays (Saasta-
moinen 1973). e statistics of the differences between
the PWV values from the radiosonde stations and those
derived from GNSS are reported in Table1. For both
radiosonde sites, the
PWVgoGPS
estimates show a bias
of less than 1mm with respect to
PWVradiosonde
, with a
standard deviation of less than 2mm, in agreement with
other comparisons between GNSS and radiosondes avail-
able in the literature (Van Baelen etal. 2005; Fujita etal.
2008; Sato etal. 2013; Realini etal. 2014).
However, the values of PWV derived from
PWVgoGPSw
exhibit a wet bias about 0.5mm larger than
PWVgoGPS
,
while the standard deviation is very similar. is could
be due to the worse parameter estimates in the GNSS
weighted adjustment, as reported in (Rothacher and
Beutler 1998). is bias does not affect the results
Fig. 6 Time variations in
PWVgoGPS
(a) and
PWVgoGPSw
(b). Each row of the matrix shows one station, identified by its four-letter code, and the
stations are ordered from west to east. Time variation interval: 10 min. The color scale shows PWV variations in (mm), where increases in PWV are
shown in yellow and decreases are shown in blue

Page 9 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Fig. 7 In the subfigure(s), the BIEL station
PWVgoGPS
and
PWVgoGPSw
time series. On the bottom, radar images corresponding to a the epoch when
the precipitation starts passing over the station, b to the epoch when it completely surrounds the station, and c to the epoch when it starts leaving
the station
Table 1 Statistics of the differences between the goGPS-derived and radiosonde-derived PWV values
Stations Distance (km)
PWVgoGPS
−
PWVradiosonde
PWVgoGPSw
−
PWVradiosonde
Mean (mm) Std (mm) Mean (mm) Std (mm)
SAVI-16113 10.9 0.47 1.47 0.90 1.55
MILA-16080 6.6 0.91 1.73 1.67 1.84

Page 10 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
presented here, which are based on the variations in ZTD
values between different epochs.
Bernese was run on a subset of the dataset (July
21–23, 2016). e comparison between
PWVgoGPS
and
PWVBernese
shows a very small bias and a standard
deviation of about 2 mm (Table2). e application of
elevation-dependent weighting also introduces a bias for
PWVgoGPSw
in this case, with a magnitude ranging from
0.4 to 2.4mm.
PWV time variation analysis
Figure6 reports the time variation of the
PWVgoGPS
and
PWVgoGPSw
time series estimated for the SPIN network
stations falling within the July 22, 2016, storm area. A
time interval of 10min was chosen to enhance detection
of the decrease in PWV following the rain event. is
decrease lasts for about an hour. If the time interval were
shorter, higher frequency variations, which are most likely
due to spurious fluctuations, would also be captured. In
Fig.6, each row of the matrix is related to one station,
and the stations are ordered from west to east in order to
highlight PWV time variations associated with the pas-
sage of the storm front. e storm front passed over the
region of interest from about 6:00 to 14:00 LT. Figure6
shows a sequence of rapid PWV decrements for each sta-
tion (blue color in Fig.6), starting from OSTA at 6:00 LT
and reaching SONP just after 12:00 LT. is behavior is
more clearly visible in the case of
PWVgoGPSw
; this is likely
due to the
PWVgoGPS
time series being smoothed by aver-
aging among all the satellite-receiver slant delays, which
are equally weighted. e satellite elevation-dependent
weighting used to obtain
PWVgoGPSw
, instead weights the
slant delays closer to the zenith direction more heavily,
increasing the temporal resolving capability of the GNSS
estimation algorithm, and thus enhancing its ability to
detect PWV fluctuations. e behavior of the PWV varia-
tions for the two easternmost stations, DARF and BORM,
appears to be uncorrelated with the other stations.
To obtain a clearer understanding of the PWV behavior
as the storm front passes, the PWV time series of BIEL,
BUSL, COMO, and NOVR stations for the July 22, 2016,
event are shown in Figs.7, 8, 9, and 10, along with the three
radar images of precipitation intensity corresponding to
the epoch at which the rain event starts passing over the
stations [labeled with (a)], the epoch when the rain event
is completely surrounding the stations [labeled with (b)],
and the epoch when the rain event has passed the stations
[labeled with (c)]. e comparison between PWV time
series and radar images highlights that the passage of the
rain front leads to a peak in PWV as the heavier precipita-
tion radar echoes surround the examined stations. Subse-
quently, a steep decrease of 5–10mm over about an hour
is evident. is decrease ends after the front has moved
out of range. e PWV behavior preceding the rain event
varies from station to station and does not display a clearly
identifiable pattern, other than a general increasing trend in
the PWV values. is increase could be caused by the fact
that by its nature, ZWD (and thus PWV as well) weights
the delays of all the satellites in view without considering
the anisotropy generated by the arrival direction of the rain
front; depending on whether satellites are actually available
in that direction, and on the number of available satellites,
the PWV increment before the rain event could be more
(or less) strongly emphasized. e PWV time variations
associated with widespread meteorological events are effec-
tively detected by a network of GNSS receivers deployed at
the regional scale, i.e., with inter-station distances on the
order of tens of kilometers. We note that the present study
does not state that the analysis of GNSS-derived PWV
time series alone is sufficient to detect rain events; a steep
decrease in these values can occur in the absence of a rain
event. Rather, our analysis shows that when a rain event
does occur, it affects the PWV values.
Table 2 Mean and standard deviation of the difference
between goGPS-estimated PWV and the Bernese-esti-
mated PWV for each station
Station
PWVgoGPS
−
PWVBernese
PWVgoGPSw
−
PWVBernese
Mean (mm) Std (mm) Mean (mm) Std (mm)
BIEL 0.2 2.5 1.0 2.2
BORM − 0.2 2.3 1.5 2.0
BRES 0.2 1.9 1.2 1.7
BUSL 0.3 2.8 1.7 2.1
CANL − 0.1 2.2 0.9 1.6
CHIA 2.2 2.3 2.4 2.1
COMO 0.2 1.8 1.1 2.0
CREA 0.0 2.3 0.9 1.9
CREM 0.1 2.6 0.9 1.8
CRSN 0.1 2.3 0.6 1.9
CUOR − 0.3 2.6 0.5 2.1
CURN 0.2 2.2 1.2 1.8
DARF 0.4 2.4 1.4 2.3
DOMS 0.7 2.7 1.7 2.2
LECC 0.5 2.3 1.4 2.0
MANT 0.4 1.8 1.2 1.6
MILA 0.0 2.4 0.9 1.7
MONV − 0.1 2.3 0.4 2.0
NOVR − 0.2 1.9 0.9 1.8
OSTA 0.3 2.9 1.4 2.3
PAV I − 0.3 2.1 0.9 1.7
SAVI 0.1 2.1 0.8 1.6
SONP 0.4 2.4 1.5 1.8
TORI 0.4 2.4 1.0 2.0
VARZ 0.4 1.8 1.1 1.6

Page 11 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Based on the results obtained in the first case study, the
second event on July 26, 2016, was analyzed by means of
PWVgoGPSw
time series alone, since they were found to
be better at detecting PWV time variations. Furthermore,
the 1-mm wet bias does not affect the characterization of
local water vapor fluctuations associated with the localized
convective phenomenon which is the object of this study.
Figure11 shows the
PWVgoGPSw
time series for the three
geodetic stations in the vicinity of the convective precipita-
tion, the time evolution of which is reported in the radar
images in the same figure. MILA and COMO, the closest
stations belonging to the SPIN regional network, are too far
apart to effectively monitor the PWV variations associated
with the local heavy rain event, and thus, their time series
do not show significant fluctuations during the time of
interest (from 15:00 to 16:00 LT). However, CATU station
(belonging to NetGeo network), although being only par-
tially affected by the convective rainfall, captures a strong
fluctuation of PWV during the time of interest. Contrary
to the July 22, 2016, event, the PWV experiences first a
steep decrease, followed by a similarly steep increase. is
shows clearly that, without the presence of CATU station,
the SPIN receivers would not have been able to observe the
convective rain cell that was the object of the study.
Fig. 8 In the subfigure(s), the BUSL station
PWVgoGPS
and
PWVgoGPSw
time series. On the bottom, the three radar images corresponding to a the
epoch when the precipitation starts passing over the station, b to the epoch when it completely surrounds the station, and c to the epoch when it
starts leaving the station

Page 12 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
e PWV time series of the low-cost GRED receiver,
the only GNSS station available within the July 26, 2016,
rainfall area, was compared to the PWV series of CATU
station (Fig.12). It is worth noting that, in this case, it
was necessary to perform the preliminary step of iono-
spheric delay interpolation from the surrounding three
stations of COMO, NOVR, and MIL to compensate for
the lack of a second frequency for the low-cost L1-only
receiver. Figure 12 shows that CATU and GRED sta-
tions, which are separated by about 8km, show similar
behavior, especially before the convective rain event.
However, further analyses are required to understand the
abnormal increasing trend in the PWV values of GRED
station beginning around 17:00 LT. is steep increase
corresponds to the time period where a decrease in PWV
would have been expected, due to the extinction of the
convective storm. Further developments of the goGPS
algorithms used to process data from low-cost receiv-
ers are planned to avoid abnormal behaviors such as that
shown in Fig. 12. A further experimental activity was
performed comparing a low-cost receiver to a geodetic
station. By applying the SEID approach to deal with the
Fig. 9 In the subfigure(s), the COMO station
PWVgoGPS
and
PWVgoGPSw
time series. On the bottom, radar images corresponding to a the epoch
when the precipitation starts passing over the station, b to the epoch when it completely surrounds the station, and c to the epoch when it starts
leaving the station

Page 13 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
ionospheric delay in both cases, the PWV values derived
from the analysis of the low-cost receiver were compared
to those obtained from the analysis of a co-located geo-
detic dual-frequency receiver that mimics an L1-only
station. is comparison was made using the data from
a period of 20 days in April 2017, a different time period
from that considered in the rest of the paper, as the geo-
detic station was not available in July 2016. We first vali-
date the PWV values obtained from the analysis of the
L1-only data from the geodetic station against the PWV
values obtained from an iono-free combination to assess
the error due to the SEID approach. is error amounts
to 0.4mm ± 0.3mm for the entire period. en, we com-
pare the PWV values of the geodetic (L1-only) and low-
cost stations, obtaining an average difference of 0.5mm
with a standard deviation of 0.4mm for the entire time
period under study. A representative 1-week subset of the
compared time series is shown in Fig.13.
Fig. 10 In the subfigure(s), the NOVR station
PWVgoGPS
and
PWVgoGPSw
time series. On the bottom, radar images corresponding to a the epoch
when the precipitation starts passing over the station, b to the epoch when it completely surrounds the station, and c to the epoch when it starts
leaving the station

Page 14 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Fig. 11 July 26, 2016, event
PWVgoGPSw
values of the stations affected by the rain event: MILA, CATU, and COMO
Fig. 12 July 26, 2016, event
PWVgoGPSw
values for CATU and GRED stations

Page 15 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Conclusions
e objective of this work was to evaluate to what extent
PWV time variations associated with heavy rain events
could be successfully captured by existing regional GNSS
networks in northern Italy. Two intense precipitation
events were studied: one characterized by a large rain
front moving from west to east, involving the north-
ern half of both Piedmont and Lombardy regions, and
a second one characterized by sparse convective cells
a few kilometers in extent. A clear pattern was found in
the analyzed PWV time series corresponding to the first
rain event studied: the PWV decreased steeply by about
5–10mm after the passage of the rain front for all the sta-
tions within the rainfall area. e analysis of the second
rain event clearly demonstrates the need for higher spatial
resolution water vapor measurements. Dense networks of
GNSS receivers could provide the required observations.
is was experimentally verified by comparing the PWV
derived from 3 geodetic GNSS receivers. is compari-
son showed that convective, localized rain phenomena do
not affect PWV time series derived from GNSS receivers
which, even if close to the phenomenon, are not directly
impacted by it. As a last step, in the framework of the
second rain event experiment, the PWV values obtained
from a single-frequency GNSS receiver were compared
to those of a nearby geodetic station. Both stations were
directly impacted by one of the localized rain cells. e
overall behavior of the low-cost PWV time series agrees
with one of the geodetic stations for the period affected
by the heavy rain; after this event, the two series display
significantly different trends, which need to be better
investigated. Although low-cost single-frequency receiv-
ers are a promising solution to the problem of densifying
GNSS networks, they still present some technological lim-
itations and require further development and calibration.
Abbreviations
GNSS: global navigation satellite systems; LT: local time; LSC: least-squares col-
location; PPP: precise point positioning; PWV: precipitable water vapor; WMO:
world meteorological organization; ZHD: zenith hydrostatic delay; ZTD: zenith
total delay; ZWD: zenith wet delay.
Authors’ contributions
SB and ER designed the study, developed the methodology, and collected the
data; SB, ER, GV, AF, and AG performed the analysis; SB, ER, and GV wrote the
manuscript. All authors read and approved the final manuscript.
Authors’ information
SB, Msc in Environmental Engineering, is a Ph.D. candidate in Geodesy and
Geomatics at Politecnico di Milano. ER, Ph.D. in Geodesy and Geomatics, is
working for the R&D company GReD, spin-off of Politecnico di Milano. GV,
Ph.D. is Associate Professor in Geodesy and Geomatics at Politecnico di Milano.
AF, Ph.D. in mathematics, is a temporary research fellow and a Ph.D. candidate
in Geodesy and Geomatics at Politecnico di Milano. AG, Ph.D. in Geodesy and
Geomatics is a temporary research fellow at Politecnico di Milano.
Author details
1 Department of Civil and Environmental Engineering, Politecnico di Milano,
Piazza Leonardo da Vinci 32, 20133 Milan, Italy. 2 Geomatics Research
and Development (GReD) srl, via Cavour 2, 22074 Lomazzo, Como, Italy.
Acknowlegements
The authors acknowledge ARPA Lombardia and ARPA Piemonte, who kindly pro-
vided the weather stations data. The authors acknowledge as well Centro Mete-
orologico Lombardo for the precipitation intensity maps, based on MeteoSwiss
radar data, and Dr. Daniele Sampietro for the interpolated temperature maps.
Competing interests
The authors declare that they have no competing interests.
Fig. 13 Comparison between PWV time series derived from a geodetic (L1-only) station and from a low-cost station for a 1-week period

Page 16 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Appendix: Hardware setup
Details concerning the hardware setup are reported in
Tables3, 4, and 5.
Table 3 Details for the Lombardy weather station network
Lombardy
ID Tow n Latitude (°) Longitude (°) Height (mm) Sensors type
Pressure Temperature Precipitation
1 Campodolcino 46.42 9.37 1880 Yes Yes Yes
2Motta Visc. 45.28 8.99 100 No Yes Yes
3Darfo B. T. 45.87 10.18 222 Yes Yes Yes
4 Laveno-Mom. 45.91 8.65 951 Yes Yes Yes
5 Canevino 44.93 9.28 455 Yes Yes Yes
6 Milano 45.47 9.22 122 Yes Yes Yes
7 Lonate Pozz. 45.59 8.74 204 Yes Yes Yes
8 Crema 45.37 9.70 79 Yes Yes Yes
9 Pavia 45.19 9.16 77 Yes Yes Yes
10 Como 45.82 9.07 201 Yes Yes Yes
11 Mantova 45.16 10.80 22 Yes Yes Yes
12 Cremona 45.14 10.04 43 Yes Yes Yes
13 Sondrio 46.17 9.88 290 Yes Yes Yes
14 Bergamo 45.70 9.67 249 No yes Yes
15 Brescia 45.51 10.22 125 Yes Yes Yes
16 Valdisotto 46.46 10.34 2295 Yes Yes Yes
17 Molteno 45.78 9.31 278 Yes Yes Yes
Table 4 Details for the Piedmont weather station network
Piedmont
ID Tow n Latitude (°) Longitude (°) Height (m) Sensors type
Pressure Temperature Precipitation
1 Alessandria 44.94 8.70 90 Yes Yes Yes
2 Biella 45.56 8.06 405 No Yes Yes
3 Demonte 44.32 7.31 765 No Yes Yes
4 Domodossola 46.10 8.30 252 No Yes Yes
5Gavi 44.69 8.80 215 No Yes Yes
6 Marene 44.67 7.73 310 No Yes Yes
7 Mondovì 44.40 7.81 422 No Yes Yes
8 Nizza Monferrato 44.76 8.35 138 No Yes Yes
9 Novara 45.44 8.63 151 Yes Yes Yes
10 Paesana Erasca 44.68 7.26 638 No Yes Yes
11 Paruzzaro 45.75 8.51 332 No Yes Yes
12 Susa 45.14 7.05 520 No Yes Yes
13 Sparone 45.41 7.54 550 No Yes Yes
14 Torino 45.08 7.68 290 No Yes Yes
15 Verolengo 45.19 8.01 163 No Yes Yes
16 Bra 44.70 7.85 298 Yes No No
17 Torino 45.19 7.65 300 Yes No No
18 Macugnaga 45.95 7.92 2075 Yes No No

Page 17 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.
Received: 31 August 2017 Accepted: 31 January 2018
References
Askne J, Nordius H (1987) Estimation of tropospheric delay for microwaves
from surface weather data. Radio Sci 22(03):379–386
Basili P, Bonafoni S, Ferrara R, Ciotti P, Fionda E, Arnbrosini R (2001) Atmospheric
water vapor retrieval by means of both a GPS network and a microwave
radiometer during an experimental campaign in Cagliari, Italy, in 1999.
IEEE Trans Geosci Remote Sens 39(11):2436–2443
Bennitt GV, Jupp A (2012) Operational assimilation of GPS zenith total delay
observations into the Met office numerical weather prediction models.
Mon Wea Rev 140(8):2706–2719
Bonafoni S, Biondi R (2016) The usefulness of the global navigation satel-
lite systems (GNSS) in the analysis of precipitation events. Atmos Res
167:15–23
Dach R, Lutz S, Walser P, Fridez P (2015) BERNESE GNSS software version 5.2.
University of Bern, Bern Open Publishing
Bevis M, Businger S, Herring TA, Rocken C, Anthes RA, Ware RH (1992) GPS
meteorology: remote sensing of atmospheric water vapor using the
global positioning system. J Geophys Res Atmos 97(D14):15787–15801
Boccolari M, Fazlagic S, Frontero P, Lombroso L, Pugnaghi S, Santangelo R,
Corradini S, Teggi S (2002) GPS Zenith total delays and precipitable water
in comparison with special meteorological observations in Verona (Italy)
during MAP-SOP. Ann Geophys 45(5):599–608
Böhm J, Niell A, Tregoning P, Schuh H (2006) Global mapping function (GMF):
a new empirical mapping function based on numerical weather model
data. Geophys Res Lett 33(7):L07304
Caldera S, Realini E, Barzaghi R, Reguzzoni M, Sansò F (2016) Experimental
study on low-cost satellite-based geodetic monitoring over short base-
lines. J Surv Eng 142(3):04015016
Cheng S, Perissin D, Lin H, Chen F (2012) Atmospheric delay analysis from GPS
meteorology and InSAR APS. J Atmos Solar Terr Phys 86:71–82
Table 5 Details for the GNSS network
Label Tow n Latitude (°) Longitude (°) Height (m) Network Receiver Antenna
ASLN Alessandria 44.92 8.62 147 SPIN LEICA GRX1200 LEIAR25.R3 LEIT
BIEL Biella 45.56 8.05 480 SPIN LEICA GRX1200 LEIAR25.R3 LEIT
BORM Bormio 46.47 10.36 1.263 SPIN TPS NETG3 TPSCR3_GGD CONE
BRES Brescia 45.56 10.23 225 SPIN TPS NETG3 TPSCR3_GGD CONE
BUSL Bussoleno 45.14 7.15 496 SPIN LEICA GRX1200 LEIAR25.R3 LEIT
CANL Canelli 44.72 8.29 206 SPIN LEICA GRX1200 LEIAR25.R3 LEIT
CHIA Chiavenna 46.32 9.40 392 SPIN TPS NETG3 TPSCR3_GGD CONE
COMO Como 45.80 9.10 292 SPIN TPS NET-G5 TPSCR.G3 TPSH
CREA Crema 45.35 9.69 130 SPIN TPS ODYSSEY_E TPSCR3_GGD CONE
CREM Cremona 45.15 10.00 103 SPIN TPS ODYSSEY_E TPSCR3_GGD CONE
CRSN Crescentino 45.19 8.11 212 SPIN LEICA GRX1200 LEIAR25.R3 LEIT
CUOR Cuorgnè 45.39 7.65 483 SPIN LEICA GRX1200 LEIAR25.R3 LEIT
CURN Curno 45.69 9.61 298 SPIN TPS NETG3 TPSCR3_GGD CONE
DARF Darfo B. T. 45.88 10.18 283 SPIN TPS ODYSSEY_E TPSCR.G3 TPSH
DEMN Demonte 44.32 7.29 863 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
DOMS Domodossola 46.12 8.29 366 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
GOZZ Gozzano 45.75 8.43 417 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
LECC Lecco 45.85 9.40 274 SPIN TPS ODYSSEY_E TPSCR3_GGD CONE
MANT Mantova 45.16 10.79 79 SPIN TPS ODYSSEY_E TPSCR.G3 TPSH
MILA Milano 45.48 9.23 187 SPIN TPS NETG3 TPSCR3_GGD CONE
MONV Mondovì 44.39 7.83 638 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
NOVR Novara 45.45 8.61 219 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
OSTA Ostana 44.69 7.19 1309 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
VARZ Pavia 45.20 9.14 144 SPIN TPS NET-G5 TPSCR3_GGD CONE
VIGE Savigliano 44.65 7.66 380 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
SERR Serravalle S. 44.73 8.85 251 SPIN LEICA GRX1200+GNSS LEIAR25.R3 LEIT
SONP Sondrio 46.17 9.87 372 SPIN TPS NETG3 TPSCR.G3 TPSH
TORI Torino 45.06 7.66 311 SPIN TPS ODYSSEY_E LEIAR25.R3 NONE
VARZ Varzi 44.82 9.20 469 SPIN TPS ODYSSEY_E TPSCR.G3 TPSH
VIGE Vigevano 45.31 8.86 169 SPIN TPS ODYSSEY_E TPSCR3_GGD CONE
CATU Cantù 45.74 9.12 364 NetGEO TPS NETG3 TPSG3_A1
GRED Lomazzo 45.70 9.04 322 GReD u-blox LEA-6T Tallysman TW3070

Page 18 of 18
Barindelli et al. Earth, Planets and Space (2018) 70:28
Ciotti P, Di Giampaolo E, Basili P, Bonafoni S, Mattioli V, Biondi R, Fionda E, Con-
salvi F, Memmo A, Cimini D, Pacione R (2003) Vespe F (2003) MERIS IPWV
validation: a multisensor experimental campaign in the Central Italy. In:
MERIS user workshop, ESA/ESRIN, Frascati, Italy
Dach R, Lutz S, Walser P, Fridez P (2015) Bernese GNSS software version 5.2.
University of Bern. Bern Open Publishing
De Haan S (2013) Assimilation of GNSS ZTD and radar radial velocity for the
benefit of very-short-range regional weather forecasts. Q J R Meteor Soc
139:2097–2107
Deng Z, Bender M, Dick G, Ge M, Wickert J, Ramatschi M, Zou X (2009)
Retrieving tropospheric delays from GPS networks densified with single
frequency receivers. Geophys Res Lett 36(19):L19802
Faccani C, Ferretti R, Pacione R, Paolucci T, Vespe F, Cucurull L (2005) Impact
of a high density GPS network on the operational forecast. Adv Geosci
2:73–79
Ferrando I (2017) GNSS Contribution to Monitor Severe Rainfalls: an Innova-
tive Procedure for Wide and Orographically Complex Area with existing
Infrastructures. Dissertation. University of Genoa
Fujita M, Kimura F, Yoneyama K, Yoshizaki M (2008) Verification of precipitable
water vapor estimated from shipborne GPS measurements. Geophys Res
Lett 35(13):L13803
Guerova G, Jones J, Dousa J, Dick G, de Haan S, Pottiaux E, Bock O, Pacione R,
Elgered G, Vedel H, Bender M (2016) Review of the state of the art and
future prospects of the ground-based GNSS meteorology in Europe.
Atmos Meas Tech 9(11):5385–5406
Herrera AM, Suhandri HF, Realini E, Reguzzoni M, de Lacy MC (2016) goGPS:
open-source MATLAB software. GPS Solut 20(3):595–603
Hofmann-Wellenhof B, Lichtenegger H, Collins J (2001) Global positioning
system. theory and practice. Springer, Wien
JMA (2013) Outline of the operational numerical weather prediction at the
Japan Meteorological Agency. Appendix to WMO technical progress
report on the global data-processing and forecasting system (GDPFS)
and numerical weather prediction (NWP) research. http://www.jma.go.jp/
jma/jma-eng/jma-center/nwp/outline2013-nwp/index.htm. Accessed 30
Aug 2017
Kouba J, Héroux P (2001) Precise point positioning using IGS orbit and clock
products. GPS Solut 5(2):2–28
Notarpietro R, Cucca M, Gabella M, Venuti G, Perona G (2011) Tomographic
reconstruction of wet and total refractivity fields from GNSS receiver
networks. Adv Space Res 47:898–912
Oolman L (2017) Atmospheric Soundings, University of Wyoming. http://
weather.uwyo.edu/upperair/sounding.html. Accessed 12 Dec 2017
Onn F (2006) Modeling water vapor using GPS with application to mitigating
InSAR atmospheric distortions. Dissertation, Stanford University
Pacione R , Pace B , Bianco G. (2013) ASI/CGS products and services in support
of GNSS-meteorology. In: EGU general assembly conference abstracts,
vol 15
Realini E, Reguzzoni M (2013) goGPS: open source software for enhancing the
accuracy of low-cost receivers by single-frequency relative kinematic
positioning. Meas Sci Technol 24(11):115010
Realini E, Sato K, Tsuda T, Susilo Manik T (2014) An observation campaign of
precipitable water vapor with multiple GPS receivers in western Java,
Indonesia. Progress Earth Planet Sci 1(1):17
Rothacher M, Beutler G (1998) The role of GPS in the study of global change.
Phys Chem Earth 23(9–10):1029–1040
Saastamoinen J (1973) Contributions to the theory of atmospheric refraction.
Bull Géod 107(1):13–34
Saito K, Shoji Y, Origuchi S, Duc L (2017) GPS PWV assimilation with the JMA
nonhydrostatic 4DVAR and cloud resolving ensemble forecast for the
2008 August Tokyo metropolitan area local heavy rainfalls. Data assimila-
tion for atmospheric, Oceanic and Hydrologic Applications, vol III, pp
383–404
Sato K, Realini E, Tsuda T, Oigawa M, Iwaki Y, Shoji Y, Seko H (2013) A high-
resolution, precipitable water vapor monitoring system using a dense
network of GNSS receivers. J Dis Res 8(1):37–47
Shoji Y, Nakamura H, Iwabuchi T, Aonashi K, Seko H, Mishima K, Ohtani R (2004)
Tsukuba GPS dense net campaign observation: improvement in GPS
analysis of slant path delay by stacking one-way postfit phase residuals. J
Meteor Soc Jpn 82(1B):301–314
Tammaro U, Riccardi U, Masson F, Capuano P, Boy J P (2016) Atmospheric pre-
cipitable water in Somma–Vesuvius area during extreme weather events
from ground-based GPS measurements. In: Freymueller JT, Sánchez L
(eds) International symposium on earth and environmental sciences for
future generations. International Association of Geodesy Symposia, vol
147, Springer, Cham
Teunissen PJG, Kleusberg A (eds) (1998) GPS for geodesy. Springer, Berlin
Van Baelen J, Aubagnac J, Dabas A (2005) Comparison of near-real time
estimates of integrated water vapor derived with GPS, radiosondes, and
microwave radiometer. J Atmos Ocean Tech 22(2):201–210