You are on page 1of 5

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325429291

Correlation Between Integrated Precipitable Water Vapor and Precipitated Water


During the Heavy Rainfall Event of February 2018 in the Tahiti Island (South
Pacific)

Conference Paper · July 2018

CITATIONS READS
3 583

5 authors, including:

Fangzhao Zhang Feng Peng


University of French Polynesia Chalmers University of Technology
22 PUBLICATIONS   74 CITATIONS    10 PUBLICATIONS   35 CITATIONS   

SEE PROFILE SEE PROFILE

Jean-Pierre Barriot Marania Hopuare


University of French Polynesia University of French Polynesia
286 PUBLICATIONS   3,404 CITATIONS    18 PUBLICATIONS   49 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Groundwater in volcanic islands and water and nitrogen isotopes View project

Multi-antenna GNSS attitude determination View project

All content following this page was uploaded by Fangzhao Zhang on 10 February 2020.

The user has requested enhancement of the downloaded file.


2018 ICEO&SI Conference, JULY 1-3, Hsinchu. Taiwan

PAPER No.:

Correlation Between Integrated Precipitable Water Vapor and


Precipitated Water During the Heavy Rainfall Event of
February 2018 in the Tahiti Island (South Pacific)
Fangzhao Zhang1, Peng Feng2, Jean-Pierre Barriot1, Marania Hopuare1, Lydie Sichoix1
1
Geodesy Observatory of Tahiti, University of French Polynesia, 98702 Faa’a, Tahiti, zhaosunshine@163.com, jean-
pierre.barriot@upf.pf, marania.hopuare@upf.pf, lydie.sichoix@upf.pf
2
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing,
Wuhan University, 430072 Wuhan, China, fengpengwhu@foxmail.com

ABSTRACT
This study focuses on the relationship between Integrated Precipitable Water Vapor (IPWV) estimates from Global
Positioning System (GPS) data and rainfall intensity in the tropical high volcanic Island of Tahiti, during a heavy rainfall
event that persisted for four weeks in February 2018 at the Geodesy Observatory site, in the suburbs of the capital city
(Papeete). During the event, the accumulated precipitation reached 500 mm, corresponding to +173% of the monthly
climatological normal. We investigate in this paper the causal relationship, for this particular event, between rainfall
intensity and IPWV in terms of cross-correlation and a causal convolution operator.

I. INTRODUCTION (2,000 mm/year) [9]. A similar study was done for Big Island
(Hawaii) by Foster et al. [10]. A detailed description of the data
Water vapor is a critical component of the greenhouse gases and methods is presented in Section II. Results and discussion
of the atmosphere, and is a major contributor to global weather are given in Section III. Conclusions are provided in Section IV.
and climate changes [1]. It also plays an important role in
weather forecasting [2]. The rainfall processes and related II. DATA AND METHODS
hazards, like flash-floods and landslides, are directly affected
by the amount of water vapor in the atmosphere [3]. Besides, A. Data Acquisition
water vapor is essential rainfall. Knowledge of the distribution 1) GPS Data: The Geodesy Observatory of Tahiti (OGT)
and temporal variation of atmosphere water vapor is therefore hosts a GPS receiver THTI (IGS reference) that consists of a
very important for better understanding of rainfall [4, 5]. Since TRIMBLE NETR8 receiver with an ASHTECH geodetic
Bevis et al in 1992 [6] first proposed to detect integrated L1/L2 antenna. The zenithal wet delay (ZWD) are estimated by
precipitable water vapor (IPWV, or PW for short) using Global using the hydrostatic VMF1/ECMWF model (VMF1 dry and
Positioning System (GPS) propagation delays, this type of data wet mapping functions) and the precise point positioning (PPP)
acquisition has been widely used to routinely monitor PW with approach under the Bernese GNSS Software 5.2 [11]. The
improved spatial and temporal resolutions w.r.t. the traditional ZWD are converted into PW estimates as follows [12], with a
meteorological techniques based on radiosonde (RS) and water 15 min sampling
vapor radiometer (WVR) observations [7].
PW = Π × ZWD (1)
The objective of this study is to investigate the relationship
between rainfall intensity and PW during a heavy rainfall event where
in Tahiti that occurred from January 29 to February 24, 2018 at
the Geodesy Observatory site, in the suburbs of the capital city 106
Π= (2)
(Papeete), at the core of the wet season. The wet season is 𝑘 ′
𝜌𝑅𝑣 ( 3 + 𝑘2 )
characterized by alternating sunny and rainy days, generated by 𝑇𝑚
the increased convective activity within the South Pacific where  is a conversion coefficient,   1000 kg / m 3 is the density
Convergence Zone [8], or by eastward propagating
perturbations. The rainfall events are often inducing intense of liquid water, R v  461 . 495 J / kg  K is the specific gas
flash-floods and landslides, claiming many lives each year. The constant of water vapor, k 2  22 . 1 K / mb
'
and
spatial rain distribution shows a large contrast between the k 3  3 . 739  10 K
5 2
are physical constants [13]. The
/ mb
northeastern part of Tahiti, wetter and exposed to dominant
trade winds (4,000 mm/year), and the southwestern part, dryer weighted average of the atmospheric temperature T m is a key
and protected from the winds by the large orographic relief parameter in this conversion process. In our case, the Tm model
2018 ICEO&SI Conference, JULY 1-3, Hsinchu. Taiwan

PAPER No.:

used is from Zhang et al. [13] as 𝑇𝑚 = 4.36𝑇𝑠 − 979.71 (wet transported by winds over large distances, and it is safer to say
season model). The local surface temperature Ts is from an that Eq. (5) only codes a partial causality between PW and
ancillary temperature sensor collocated with the rain gauge that rain.
acquired our rain data.
An additional difficulty, in both the cross-correlation and
2) Rain Gauge Data: The rain gauge station, with a 1 the convolution modeling is that the spectral contents of the
min sampling, is collocated with the GPS station THTI. The time series of the rain and of the PW are quite different. The
recorded data includes temperature, relative humidity, pressure, rain records are basically a series of “spikes”, and the PW
rainfall accumulation and rainfall intensity. records are much smoother (see Fig. 1-a). Besides, to have a
rainfall event implies that the PW must reach a threshold (the
B. Methods rain is not a continuous function of the PW), which has also to
We calculated and analyzed the correlation and convolution be determined in some way from Eq. (5). We introduce this
between PW estimates from GPS and rainfall intensity in order threshold 𝜉 as
to investigate their relationship. 𝑡

1) The cross-correlation between PW and Rainfall Rain(t) = ∫ 𝐾(𝑡 − 𝑧) (𝑃𝑊(𝑧) − 𝜉) 𝑑𝑧 (7)


−∞
Intensity is defined as
We will estimate both a discretized form (with a one min
+∞
sampling) of the kernel K and the threshold 𝜉 through a non-
R(T) = ∫ 𝑃𝑊(𝑡 + 𝑇) 𝑅𝑎𝑖𝑛(𝑡)𝑑𝑡 (3)
−∞
linear least-squares fit (as K and 𝜉 are multiplicative) from the
rain data [18].
Where T is the phase shift, and can be written in a normalized
form as
III. RESULTS AND DISCUSSION
∑𝑖 𝑃𝑊(𝑡𝑖+𝑗 ) × 𝑅𝑎𝑖𝑛(𝑡𝑖 ) A heavy rainfall event occurred in Tahiti, French Polynesia,
R(j) = (4)
√∑𝑖 𝑃𝑊 2 (𝑡𝑖 ) √𝑅𝑎𝑖𝑛2 (𝑡𝑖 ) during four weeks in February 2018. The precipitation totals
exceeded 500 mm at the Geodesy Observatory of Tahiti (OGT)
where R (j) is the correlation coefficient, with i = 0,1, … , N site located at 100 m altitude ( 149°25′ W, 17°40′S ).
and j = 0,1, … , M. Simultaneously, a large accumulated rainfall of 572 mm was
2) We write the causal convolution between PW and confirmed at the Faa’a Airport station at near sea level altitude
rainfall intensity as operated by Météo-France Weather Centre (about 1.6 km to the
north of OGT).
𝑡
Rain(t) = ∫ 𝐾(𝑡 − 𝑧) 𝑃𝑊(𝑧) 𝑑𝑧 (5) A. Correlation of PW with Rainfall Intensity
−∞
Fig. 1-a depicts, for this heavy rainfall event, the PW time
where K is the causal (i.e. with zero value when z > t) series (with a 1 min linear sub-interpolation) and the variations
convolution kernel, or in other words of rainfall intensity. During this event, the PW values were
larger (up to 75 mm) than usual (around 50 mm). The rain
Rain[i] = ∑ 𝐾(𝑖 − 𝑗) ∗ 𝑃𝑊[𝑗] (6) began with the first increase in PW and ended with a sharp
𝑗 decrease in PW. Figure 1-b shows that there is only a weak
correlation between the humidity on the ground level and the
with i = 0,1, … , N and j = 0,1, … , M. A similar approach can
PW. Fig. 2-a shows the evolution of the correlation coefficient
be found in [14]. R is as a function of the phase shift T. One can see clearly in
By using a causal kernel, we assume that the PW is the Fig. 2-b a phase shift of 16 min between PW and rain data,
ultimate source of rain. Albeit that is indisputably true, the rain indicating that, in the mean, the rain starts 16 min after a
is also a source of PW through evaporation and maximum of PW. Another “bump” can be seen on the right of
evapotranspiration by the plants [15]. In other words, Eq. (3) is the curve in Fig. 2-a at a phase shift of 6 hours, probably
relative to the whole loop of water between the surface and the related to the evapotranspiration process that is feeding up the
atmosphere (rain and evaporation/evapotranspiration), and Eq. PW. Serafini et al. [9] found that a phase shift of half-an-hour
(5) only focus on the rain process. We have also to stress that between the PW and rain for Tahiti Island over eight-year
the path from PW to rain is a complicated one, where the period. The difference with our value is probably due to three
nucleation of excess water vapor into cloud droplets is induced factors: the rain gauge was not collocated with the GPS
by external agents like aerosols (saline aerosols in our case receiver (7 km distance); they did not separate the wet and dry
[16]), and/or even cosmic rays [17]. Besides, clouds can be seasons; and of course their study spanned eight years of data.
2018 ICEO&SI Conference, JULY 1-3, Hsinchu. Taiwan

PAPER No.:

(a) (b)

Figure 1. (a) Time series in February 2018 of 1/ estimates of PW based on the wet season Tm model with 1 min linear interpolation (blue dots) and 2/ collocated
rain intensity (red dots), and (b) Relationship between PW and ground relative humidity (the humidity sensor seems to saturate for values higher than 94 %).

(a)

(b)

Figure 2. (a) Evolution of linear correlation coefficient R as a function of the phase shift T (Eq. (4)) between the time series of PW and collocated rain intensity,
and ((b) partial enlargement of (a)): the phase shift between the time series of PW and rain intensity is 16 min (red line). One can also see, in the right part of sub-
Fig. (a) a "bump" at around 6 h of phase shift that is probably related to the evapotranspiration processes occurring after a rain event.

B. Convolution of PW with Rainfall Intensity


To model the causal relationship between the PW and the reconstructed rain in the vertical direction. We found a “best”
rain is a challenge, as their frequency contents are very value of 48.2 mm for 𝜉 . In their study about a storm over Big
different. To overcome this difficulty, we used an inverse Island, Foster et al. found a value of 53 mm (figure 7 of
problem approach, with a regularization done by only reference [10]), but their orographic context is very different.
considering the statistically meaningful singular values of the We stress again, the cross-correlation phase shift of 16 min
kernel K [18]. The reconstructed kernel is shown in Figure 3-a, and the convolution maximum occurring at 29 min does not
with a maximum occurring at 29 min. We adjusted (Fig 3-b) represent the same physical meaning. The first one integrates
the threshold value 𝜉 by fitting “by eye” the recorded rain (in the whole “water loop” in the atmosphere, and the second one
red), to the reconstructed rain (in green), as we noticed that the isolates the causal relationship between the PW maxima and
variation of 𝜉 was basically inducing a shift of the the rainfall events.
2018 ICEO&SI Conference, JULY 1-3, Hsinchu. Taiwan

PAPER No.:

(a) (b)

Figure 3. (a) Evolution of convolution kernel (Eq. 6) between PW and rain intensity, with a maximum occurring at 29 min (red line), and (b) Time series in
February 2018 of rain intensity from rain gauge (red dots) and reconstructed rain intensity (green line), that is closely an image of PW (see Figure1-a).

[7] Ohtani R, Naito I. Comparisons of GPS-derived precipitable water vapors with


IV. CONCLUSIONS radiosonde observations in Japan [J]. Journal of Geophysical Research
In this paper, we have analyzed and compared the Atmospheres, 2000, 105(D22):26917-26929.
relationship between PW and rainfall intensity during a heavy [8] Haffke C, Magnusdottir G. The South Pacific Convergence Zone in three decades
of satellite images[J]. Journal of Geophysical Research Atmospheres, 2013,
rainfall event of February 2018 in the island of Tahiti (South 118(19):10,839–10,849.
Pacific). We demonstrated that this relationship can be
[9] Serafini J, Barriot J P, Sichoix L. The evolution of precipitable water and
characterized, in the mean, by three numbers: a phase shift of precipitation over the Island of Tahiti from hourly to seasonal periods [J].
16 min, characterizing the whole “water loop” in the International Journal of Remote Sensing, 2014, 35(18):6687-6707.
atmosphere, a causal 29 min delays between the maxima of the [10] Foster, J., M. Bevis, Y.-L. Chen, S. Businger, and Y. Zhang, The Ka‘u storm
PW and the rainfall events, and a threshold of 48.2 mm for the (November 2000): Imaging precipitable water using GPS, J. Geophys. Res.,
PW to induce rain. This study provides further evidence that 108(D18), 4585, doi:10.1029/2003JD003413, 2003.
the PW estimates derived from GPS data can be assimilated in [11] Boehm J, Werl B, Schuh H. Troposphere mapping functions for GPS and very long
rain forecasting and global climate change models. baseline interferometry from European Centre for Medium ‐ Range Weather
Forecasts operational analysis data[J]. Journal of Geophysical Research Solid
Earth, 2006, 111(B2):-.
ACKNOWLEDGMENT
[12] Davis J L, Herring T A, Shapiro I I, et al. Geodesy by radio interferometry: Effects
This research was funded by the French Space Agency of atmospheric modeling errors on estimates of baseline length [J]. Radio Science,
(CNES) and the University of French Polynesia (UPF). 1985, 20(6):1593–1607.
[13] Zhang F, Barriot J-P, Xu G, Yeh T-K. Metrology Assessment of the Accuracy of
REFERENCES Precipitable Water Vapor Estimates from GPS Data Acquisition in Tropical Areas:
The Tahiti Case. Remote Sensing. 2018; 10(5):758.
[1] Duan J M. GPS meteorology: Direct estimation of the absolute value of
[14] J.-P. Barriot, L. Pheulpin, Lydie Sichoix, Su-Min Shen, Mean Time Response
precipitable water [J]. Journal of Applied Meteorology, 1996, 35(6):830-838.
Curve of a Small Watershed, in Tahiti, French Polynesia, 2016 ICEO&SI
[2] Yeh T K, Hong J S, Wang C S, et al. Determining the precipitable water vapor with Conference, June 26-28, Keelung. Taiwan.
ground-based GPS and comparing its yearly variation to rainfall over Taiwan [J].
[15] Jasechko S., Sharp Z. D., Gibson J.J., Birks S.J., Yi Y. and P.J. Fawcett, Terrestrial
Advances in Space Research, 2016, 57(12):2496-2507.
water fluxes dominated by transpiration, Nature, 496, 347,
[3] Sharifi M A, Khaniani A S, Joghataei M. Comparison of GPS precipitable water doi:10.1038/nature11983, 18 April 2013.
vapor and meteorological parameters during rainfalls in Tehran[J]. Meteorology &
[16] O'Dowd C.D. and G. De Le Leeuw, Marine aerosol production: a review of the
Atmospheric Physics, 2015, 127(6):701-710.
current knowledge, Phil. Trans. R. Soc. A, Vol. 365, pp 1753–1774,
[4] Fadil A, Sari D B. Correlation Between Precipitable Water and Rainfall Using doi:10.1098/rsta.2007.2043, 2007.
Global Positioning System (GPS) Technique [J]. Nato Security Through Science,
[17] Svensmark H., Enghoff M. B. and J. O. P. Pedersen, Response of cloud
2006:271-284.
condensation nuclei (>50 nm) to changes in ion-nucleation, Physics Letters A, Vol.
[5] Wu P, Mori S, Hamada J I, et al. Diurnal Variation of Rainfall and Precipitable 377(37), pp 2343-2347, doi:10.1016/j.physleta.2013.07.004, 2013.
Water over Siberut Island off the Western Coast of Sumatra Island [J]. SOLA -
[18] Menke W, Geophysical Data Analysis: Discrete Inverse Theory, Academic Press,
Scientific Online Letters on the Atmosphere, 2008, 4(1):125-128.
Vol 45, Int. Geophysics Series, 1989.
[6] Bevis, Businger, T.A, et al. GPS meteorology - remote sensing of atmospheric
water vapor using the Global Positioning System[C]// Global Engineering
Education Conference. IEEE, 1992:844-846.

View publication stats

You might also like