Earthquakes_and_heavy_rainfall_influence_on_aquife

RESEARCH ARTICLE Earthquakes and Heavy Rainfall Influence on Aquifer
10.1029/2022WR033367
Properties: A New Coupled Earth and Barometric Tidal
Key Points:
• A model for the tidal response of
Response Model in a Confined Bi-Layer Aquifer
bi-layer aquifers was developed, A. Thomas1 , J. Fortin1 , B. Vittecoq2 , and S. Violette1,3
focusing on pressure diffusion and
exchange between two confined layers 1
Laboratoire de Géologie, Ecole Normale Supérieure - PSL Research University, CNRS, UMR8538, Paris, France, 2BRGM,
• Using 14 years of hourly piezometric
Fort-de-France, Martinique, 3Sorbonne University, UFR.918, Paris, France
data, the temporal evolution of aquifer
diffusivity was inferred and confirmed
independently by pumping tests
• The change of permeability is shown Abstract Among the impacts of earthquakes on aquifers, permeability change is one of the most
to be due to earthquakes, extreme challenging to quantify, since techniques to measure permeability evolution are scarce. The study of tidal
rainfall events and aquifer withdrawals
in upstream borehole
response of boreholes is one of the most promising, yet complex to use in practice. We used 14 years of
piezometric level measurements and two concurrent source signals, earth tidal strain and barometric pressure,
for which we separated the respective contribution in a state-of-the-art tidal analysis. We developed a new
Supporting Information:
general analytical hydrogeological model, based on geological observations of a confined bi-layer aquifer. It
Supporting Information may be found in
the online version of this article. is able to match combined observations of earth and barometric tide phase lags which could not be explained
by existing models. We demonstrate that its relative complexity can be overcome thanks to the results of
Correspondence to:
tidal analysis, yielding a simpler model adapted to the Fond Lahaye site of the Martinique Island. The
A. Thomas, resulting evolution of diffusivity and loading efficiency, was validated independently with several pumping
augustin.thomas@ens.fr tests occurring all along the studied period. The transient diffusivity increases and decreases indicate which
earthquakes impacted the aquifer, enabling to establish an empirical magnitude-distance relationship criterion.
Citation: This criterion confirms the suspected dependence on dynamic stresses, which decrease as the square of the
Thomas, A., Fortin, J., Vittecoq, B., hypocentral distance. Additionally, we investigate two other factors of diffusivity changes: heavy rainfall events
& Violette, S. (2023). Earthquakes and aquifer withdrawals, which demonstrates the sensitivity of volcanic aquifers properties to environmental
and heavy rainfall influence on
aquifer properties: A new coupled and anthropogenic influence.
Earth and barometric tidal response
model in a confined bi-layer aquifer.
Water Resources Research, 59, 1. Introduction
e2022WR033367. https://doi.
org/10.1029/2022WR033367 Piezometric level response to periodic forcings, like solid earth tide strain, barometric loading or oceanic tide
loading, is a useful alternative way to characterize aquifer properties. It was first spread as a tool for aquifer
Received 4 AUG 2022
Accepted 6 MAR 2023
characterization 34 years ago with the model of Hsieh et al. (1987). It was promising in the sense it only needs
classical monitoring data (hourly water level) and no expensive field work like a pumping test. Yet its usage is
still far from its potential as expressed in the review of McMillan et al. (2019). On the one hand, the pumping test
literature, relatively old, covers a wide variety of aquifer geometries and boundary conditions, including, without
comprehensiveness, models for confined radial aquifer (Theis, 1935), leaky aquifers (Hantush & Jacob, 1955),
double porosity aquifers (Warren & Root, 1963), unconfined aquifer (Neuman, 1972), spatially heterogeneous
aquifers with the general radial flow model (Barker, 1988). Research is still ongoing namely on numerical mode-
ling to complexify the possible responses. On the other hand, literature on periodic responses only recently repre-
sented a decent variety of aquifer geometry, starting from Roeloffs (1996) and Rojstaczer (1988) for unconfined
aquifers, to Wang et al. (2018) for leaky aquifers. Research on this topic now faces the challenge of proposing a
comprehensive view and practical guidelines, like did in its time the useful tutorial of Doan et al. (2008). Now
the two major and somewhat opposing pitfalls scientists face are the following. The first one is that the existing
models are based on strong hypotheses (perfect confinement for Hsieh et al. (1987), negligible storage in the
aquitard in Wang et al. (2018),) which of course are not in general met. There is still room for a more general
derivation of analytical models, like what Odling et al. (2015) did in the case of the model of Rojstaczer (1988),
precising the low frequency behavior which was improperly predicted. The second pitfall, opposing to the first, is
that we need to remain as simple as possible, and keep in mind the mathematical limitations of modeling inver-
sion: from n observed variables, we can invert at best n parameters. In that sense, periodic response problems are
nearly always over-parametrized, since observations are often limited to phase and amplitude response at one or
© 2023 American Geophysical Union. two frequencies, while hydrodynamic and poroelastic parameters (hydraulic conductivity, storativity, etc.), grow
All Rights Reserved. numerous with the complexity of the models. In this paper, we do not claim to give the comprehensive view we
THOMAS ET AL. 1 of 26
Water Resources Research 10.1029/2022WR033367
call for, but go one step toward and present a case study where we try to methodically avoid the two previously
identified pitfalls. Building on the geological knowledge of the field (Section 2.1), we build a new model adapted
to our aquifer geometry, without a priori hypotheses on the parameters (Sections 4.1–4.3). To better the odds
of being able to invert the model, we calculated the tidal responses of the two significant signal sources, and
disentangled their contributions (Section 3), like recommended in McMillan et al. (2019), and implemented in
Valois et al. (2022). As a secondary step only, we use all of this information to constrain and invert the model
(Sections 4.5 and 4.6). Thanks to a validation with several pumping tests (Section 4.7), we can confidently
discuss the evolution of the inverted parameters and the different phenomena at stakes (Section 5.1). Indeed,
a major benefit of the method is to provide the evolution of hydrodynamic parameters evolution across time.
Thus, it was frequently implemented to study the response of aquifers to earthquakes (Elkhoury et al., 2006; Shi
et al., 2019; Sun et al., 2020; Wang et al., 2016; Zhang, 2019), which where proven to have numerous impacts
on aquifer systems, from water level oscillations, water level coseismic drop or rise (Brodsky et al., 2003), new
springs appearance or increase in stream discharges (Wang & Manga, 2015), mud volcano eruptions (Manga &
Wang, 2015), to what interest us particularly here: permeability changes (Elkhoury et al., 2006; Ingebritsen &
Manga, 2019; Rojstaczer et al., 1995). Several processes have been proposed in the literature to explain perme-
ability changes, including effective pressure changes (Muir-Wood & King, 1993; E. Roeloffs et al., 2003),
clogging-unclogging of fractures (Barbosa et al., 2019; Brodsky et al., 2003; Candela, 2014), opening of new
fractures (Wang et al., 2016; Xue et al., 2013) or liquefaction/consolidation (Manga, 2001; Montgomery &
Manga, 2003). In the far field, these changes, caused by dynamic stresses are often transient (Manga et al., 2012),
and the empirical relationships linking the maximum distance at which an earthquake of given magnitude can
affect the crust depend on the phenomena. Roeloffs (1998) proposed a relationship for constant pore pressure
changes, Wang and Manga (2010) proposed the seismic energy density as a general metric. Here we show that
dynamic stresses may be more relevant in the far field, since they decay as the square of the distance for surface
waves (Section 5.1). Finally the time-dependent diffusivity inverted also showcases a response to heavy tropical
rainfall events like observed in another watershed in Martinique (Vittecoq et al., 2020) as well as to aquifer with-
drawals, illustrating how sensitive to environmental and anthropogenic influence fractured aquifers properties
can be.
2. The Martinique Fond Lahaye Aquifer

2.1. Geological Context
Martinique is a volcanic island of the Lesser Antilles archipelago, resulting from the subduction of the Atlantic
lithosphere bellow the Caribbean plate (Figure 1). It is the largest volcanic island of the archipelago (1,080 km 2),
with a volcanic activity, mainly andesitic, for at least 25 Ma (Westercamp et al., 1990). The resulting relief is
mountainous in the north (highest and youngest volcano, Montagne Pelée, at 1,397 m) and gentler in the south with
highest relief at 504 m. Climate in Martinique is humid tropical, with a rainy season between July and November
with precipitation brought by trade winds reaching 5,000–6,500 𝐴𝐴 mm ⋅ yr −1 at the northern summits and between
1,200 and 1,500 mm yr in the south. Temperatures vary between 18 and 32°C at Fort-de-France (Vittecoq
−1
et al., 2019). Over the period 2007–2019, 128 earthquakes were felt in Martinique (Figure 1) and reported by
the Volcanological and Seismological Observatory of Martinique (OVSM-IPGP) (min magnitude 2.1, median
4.2 and max 7.4). The Fond Lahaye borehole is situated in a valley situated on the west coast of the island, 2 km
from the sea, at an elevation of 76.3 m amsl (Figure 1). The geological formations in the valley are 5.5-2.2 Ma
fractured andesite lava flows topped by conglomerates, debris, block and alluvium from the dismantling of the
upper volcanic edifice (Carbet volcanic complex) (Vittecoq et al., 2019; Westercamp et al., 1990). The aquifer
corresponds to the fractured andesite (30–62 m below surface), which upper part has been altered up to residual
andesite blocks in a clay matrix which act as a barrier for fluid flow (Figure 2). The conglomerates and alluvium
on top (2–22 m below surface) are dry: no water inflow was observed during drilling, this is also confirmed by
resistivity measurements as shown by the resistivity log on Figure 2 (Vittecoq & Brugeron, 2008). The borehole
is 62 m deep, and opened to the aquifer between −22 and −62 m. The drilling 𝐴𝐴 radius 𝐴𝐴𝑤𝑤 is 194 mm and the casing
radius 𝐴𝐴𝑐𝑐 125 mm. The water level, which is above the aquifer top, confirms the confined type of the aquifer.
𝐴𝐴
2.2. Data Presentation

Data consists of 14 years of hourly piezometric level records, starting from February 2008 (Figure 3a). Two
sensors have been used both with millimetric resolution. OTT Thalimedes from 2005 to 2014, a float operated
Figure 1.
shaft encoder with integrated data logger, and SEBA Dipper-PT data logger,
from 2014 to 2022, a ceramic pressure sensor, with integrated air pressure
compensation tube.
Tidal signals could not be analyzed on some intervals between May 2010 and
April 2013 due to sensor malfunction. Pumpings from late 2018 in a borehole
situated 180 m upstream, with either high drawdown or on a daily rhythm,
created spurious signal in the tidal frequency band. Thus, we focus the signal
study during two timeseries: from February 2008 to April 2010 and from
May 2013 to August 2018. Data which has been rejected is signaled in dark
blue on Figure 3a. The barometric data as well as rainfall data have been
purchased from the French meteorological agency nearest station, 5.3 km
south east of Fond Lahaye (Figure 1c). Barometric data is measured hourly
with a digital barometer PTB220 – VAISALA. Rainfall data is presented on
a daily basis (Figure 3b) until early 2018, which covers the whole studied
interval. Synthetic Earth tidal strain data have been generated with the func-
tion ertid of the SPOTL software (Agnew, 2012). Areal strain was converted
to volumetric strain with an hypothesis of a Poisson ratio of 0.3 (Doan
et al., 2008, eq 2.17).
2.3. Pumping Tests

Pumping tests were conducted in 2007, 2008 (Vittecoq & Brugeron, 2008),
2013 and 2022. The three firsts correspond to long terms pumping tests
(72 hr for the two first and 42 days for the third one) conducted in a borehole
190m upstream in the Fond Lahaye valley (National number 1177ZZ0177).
The studied borehole was used as a piezometer thus the interpretation yielded
the storativities as well as hydraulic conductivities. The fourth pumping
test was a short term one (4.5 hr), conducted in the studied borehole itself,
with no piezometer. Data was analyzed with the MLU software (Carlson
& Randall, 2012), which analytically solves drawdown in a multi-layered
system. Results from pumping tests, used to validate a posteriori the model,
are presented in Section 4.7.
Figure 2. Geological borehole log. The aquifer is made of two layers: the
fracture andesite and the altered andesite. The layer “altered andesite in a clay 3. Tidal Analysis
matrix” forms a cap rock. Thus, the aquifer layers (fractured andesite + altered
andesite) are confined. 3.1. Tidal Analysis Method
The first step is to analyze the piezometric signal to compute barometric
and earth tidal response functions. Practically response functions are (a) the
amplitude ratio and (b) the phase lag between each source term (barometric and earth tide) and the response
measured in the piezometric level.
Piezometric level in the borehole can be decomposed as a trend superposed to a tidal response:
(1)
ℎ𝑤𝑤 (𝑡𝑡) = ℎ𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 (𝑡𝑡) + ℎ𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 (𝑡𝑡) .
We are interested in the tidal signal which we obtained by filtering the signal with a Butterworth bandpass filter
of order 4 with cutoff frequencies of 0.7 and 2.6 cpd (cycles per day). This signal can be written as the sum of
cosine of known frequencies as predicted by tidal theory (Agnew, 2015):
Figure 1. (a) Localization of Martinique in the Caribbean plate. Caribbean plate boundary (dark line) and approximate North/South American plate boundary (white
shading) from (Braszus et al., 2021) (b) Localization of the 128 felt earthquakes in Martinique reported by the OVSM-IPGP observatory during the period 2007–2019.
Four strongest earthquakes: M7.4–2007/11/29 (hypocentral distance—hd: 165 km), M6.5–2014/02/18 (hd: 219 km), M6.6–2015/07/16 and M5.6–2017/02/03 (hd:
91 km). Digital elevation model from the General Bathymetric Chart of the Ocean (gebco.net), White arrow: vector of convergence (DeMets et al., 2000), main faults,
ridges and subduction trench (Leclerc, 2014). (c) Localization of the Fond Lahaye Borehole, the ocean tide gauge (Fort-de-France harbor), the barometer (barometric
pressure data) and the rain gauge station. Figure performed using ArcGis 10.5.1 (https://www.esri.com/).
Figure 3. (a) Piezometric level (m below ground level) of the Fond Lahaye borehole time series (2008–2022), with a
1-month zoom on September 2009. Color indicates whether the data was used or not in the computation of the transfer
functions. (b): Rain gauge data from Fort-de-France Desaix station (Figure 1c), in mm/day (c). (d, e, f): Frequency spectrums
of piezometric level, barometric pressure, earth tidal strain and oceanic tide respectively. Each frequency is designated by
its Darwin symbol (O1: 0.9295 cpd, S1: 1.0000 cpd, K1: 1.0027 cpd, M2: 1.9322cpd, S2: 2.0000 cpd (Agnew, 2015), cpd:
cycles per day).
𝑛𝑛
(2)
∑
ℎ𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 (𝑡𝑡) = 𝐴𝐴𝑘𝑘 cos(𝜔𝜔𝑘𝑘 𝑡𝑡 + 𝜑𝜑𝑘𝑘 ) .
𝑘𝑘=1
𝐴𝐴With 𝐴𝐴𝑘𝑘 , 𝜙𝜙𝐴𝐴𝑘𝑘 and 𝐴𝐴𝑘𝑘 the angular frequency and corresponding phase and amplitude respectively. Or in complex
notation:
𝑛𝑛
(3) ℎ̃ 𝑤𝑤𝑤𝑤𝑤 ei𝜔𝜔k 𝑡𝑡 ,
∑
ℎ𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 (𝑡𝑡) =
𝑘𝑘=1
𝐴𝐴with ℎ̃ 𝑤𝑤𝑤𝑤𝑤 the complex amplitude. We recover each complex amplitude

𝐴𝐴 ℎ̃ 𝑤𝑤𝑤𝑤𝑤 through a least square fit of sines and
cosines at the known frequencies (Harmonic least square fit or HALS). Each HALS is performed on a 29 days
window, the minimum width required to distinguish the two frequencies M2 and S2 (Agnew, 2015). The same
treatment is applied to barometric pressure and earth tidal strain written as:
𝑛𝑛
(4) 𝑃𝑃̃𝑎𝑎𝑎𝑎𝑎 ei𝜔𝜔k t ,
∑
𝑃𝑃𝑎𝑎 (𝑡𝑡) =
𝑘𝑘=1
𝑛𝑛
∑
(5)
𝜀𝜀(𝑡𝑡) = 𝜀𝜀̃𝑘𝑘 ei𝜔𝜔k t ,
𝑘𝑘=1
̃𝑎𝑎 and 𝐴𝐴𝐴 the complex amplitudes. Within the framework of the linear theory of poroelasticity, we assume
𝐴𝐴with 𝑃𝑃𝐴𝐴
each frequency component of the piezometric signal is the result of the sum of effects from atmospheric pressure
and earth tidal strain:
1(
∀𝑘𝑘 ∈ ⟦1, 𝑛𝑛⟧, ℎ̃ 𝑤𝑤𝑤𝑤𝑤 =
(6) 𝑍𝑍𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 ⋅ 𝑃𝑃̃𝑎𝑎𝑎𝑎𝑎 + 𝑍𝑍𝜀𝜀𝜀𝜀𝜀 ⋅ 𝜀𝜀̃𝑘𝑘
)
𝜌𝜌𝜌𝜌
In general, we are left for each frequency with one equation and two unknowns: the transfer functions𝐴𝐴 𝐴𝐴𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 [−]
𝐴𝐴 and 𝐴𝐴𝜀𝜀𝜀𝜀𝜀 [𝑃𝑃 𝑃𝑃] which are frequency dependent. To solve this problem in the case of tidal signals, we assume the
transfer functions do not vary quickly with frequency, which was already done in Acworth et al. (2016) and more
recently in Valois et al. (2022). This hypothesis will be discussed in Section 5.3. For two close frequencies
𝐴𝐴 𝐴𝐴1
𝐴𝐴 and 𝐴𝐴2, we can write:
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤𝑤1 = 𝑍𝑍𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏12 ⋅ 𝑃𝑃̃𝑎𝑎𝑎1 + 𝑍𝑍𝜀𝜀𝜀12 ⋅ 𝜀𝜀̃1

(7)
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤𝑤2 = 𝑍𝑍𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏12 ⋅ 𝑃𝑃̃𝑎𝑎𝑎2 + 𝑍𝑍𝜀𝜀𝜀12 ⋅ 𝜀𝜀̃2
This linear system is solved as:
𝑃̃𝑎,1 ⋅ ℎ̃ 𝑤,2 − 𝑃̃𝑎,2 ⋅ ℎ̃ 𝑤,1

𝑍𝜀,12 = 𝜌𝑔
𝑃̃𝑎,1 ⋅ 𝜀̃2 − 𝑃̃𝑎,2 ⋅ 𝜀̃1
(8)
𝜀̃2 ⋅ ℎ̃ 𝑤,1 − 𝜀̃1 ⋅ ℎ̃ 𝑤,2
𝑍𝑏𝑎𝑟𝑜,12 = 𝜌𝑔
𝑃̃𝑎,1 ⋅ 𝜀̃2 − 𝑃̃𝑎,2 ⋅ 𝜀̃1
This method enables to disentangle accurately the effects of barometric and earth tidal loading in the case of
a pair of close frequencies. This is the case for the most studied semidiurnal tides M2 and S2 (Agnew, 2015)
at 1.93 and 2 cpd (cycles per day) respectively, or the diurnal couple O1 S1 at 0.93 and 1 cpd. This method is
useful to study both responses to atmospheric loading and earth tides. Studies addressing either atmospheric
loading or earth tides are numerous (Acworth et al., 2016; He, 2016; Rojstaczer & Agnew, 1989; Rojstaczer &
Riley, 1990; H. Zhang et al., 2019). Recently this method was implemented to combine both signals, namely by
Rau et al. (2022) or Valois et al. (2022).
3.2. Tidal Analysis Results
The frequency spectrums of piezometric level, barometric pressure, earth tidal strain and oceanic level are repre-
sented respectively on Figures 3c, 3d, 3e and 3f. Given the two first spectrum similarities (Figures 3c and 3d), it
appears that the piezometric level response is mostly driven by barometric pressure. Barometric pressure is also
the only source that has a S1 frequency (1.0000 cpd) which is close but distinguished from the K1 frequency
Figure 4. Evolution of the transfer functions over the last 10 years (a) Barometric and (b) Earth tide phase lags versus time. (c) Barometric gain and (d) Piezometric
response to earth tide amplitude (in mm – left - and in Pa/nstrain - right) versus time. Dark vertical lines correspond to the main earthquakes in the region (discussed in
Section 5.1).
(1.0027 cpd), presented by both oceanic and earth tides. Yet the increase in M2 amplitude relatively to S2 in
these spectrum (Figures 3c and 3d) reveals another contribution either from earth tide or oceanic tide that cannot
be neglected. If the additional contribution would come from oceanic tide (Figure 3f), a significant peak in O1
should have been observed in the piezometric level (Figure 3c), which is not the case. Thus, oceanic tide contri-
bution is neglected before earth tide; this assumption will be discussed in Section 5.3. Finally, a complete tidal
analysis as described in Section 3.1 was performed. Using M2 and S2 frequencies 𝐴𝐴 as 𝐴𝐴𝐴𝐴1 and 𝐴𝐴2, equation
𝐴𝐴 (8)
yields the two transfer functions represented on Figure 4, on which each point corresponds to a 29 days window,
and two adjacent points are 10 days apart. The barometric phase lag (Figure 4a) is defined 𝐴𝐴 as −arg 𝑍𝑍𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏2∕𝑆𝑆2
( )
( ̃∗ )
so that the phase lag is around 0° and
𝐴𝐴 not ± 180◦.This barometric phase lag corresponds 𝐴𝐴 to arg −𝑃𝑃̃𝑤𝑤 where the
𝜌𝜌𝜌𝜌 ℎ
𝑎𝑎
superscript * refers to the piezometric level response to a single source, that is, barometric loading in this case,
which is different from the measured piezometric level. We measure a negative phase lag for barometric pressure,
starting from around −10° in 2008, increasing to −5° between 2008 and 2010, seemingly to stabilize around
−3° from 2014, yet knowing some perturbations that will be discussed in Section 5.1. The amplitude response
𝜌𝜌𝜌𝜌 ℎ̃ ∗
𝐴𝐴 |𝑍𝑍𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏2∕𝑆𝑆2 |, that is, the barometric𝐴𝐴gain, | −𝑃𝑃̃𝑤𝑤 |, shows the same global trend, increasing from 0.7 in 2008 to 0.9 in
𝑎𝑎
2014 and then fluctuating around the value of 0.9, indicating that around 90% of the barometric signal is retrieved
in the piezometric level.
( ̃∗ )
The earth tide phase lag (Figure 4b) is defined 𝐴𝐴 as arg 𝑍𝑍𝜀𝜀𝜀𝜀𝜀2∕𝑆𝑆2 . This phase lag corresponds
𝐴𝐴 to arg 𝜀𝜀̃ 𝑤𝑤
( ) 𝜌𝜌𝜌𝜌 ℎ
where the superscript * refers to the piezometric level response to a single source, that is, the earth tide in this
case, which is different from the measured piezometric level. The earth tide gain used in this article will be
𝜌𝜌g̃h∗ 𝜌𝜌g̃h∗
𝐴𝐴 𝐴𝐴 = |𝑍𝑍𝜀𝜀𝜀𝜀𝜀2∕𝑆𝑆2 | = | 𝜀𝜀̃ w |, except in Figures 6 and 15 where the gain is normalized 𝐴𝐴 to 𝐴𝐴𝑛𝑛 = | 𝐵𝐵𝐵𝐵 w𝜀𝜀̃ | to facilitate
𝑢𝑢
comparison with the literature. The earth tide amplitude response (Figure 4d) is expressed either in Pa/nstrain
(unit
𝐴𝐴 of 𝐴𝐴 ), or in mm of piezometric head. In the latter case, the response is computed 𝐴𝐴 as 𝜌𝜌𝜌𝜌 |𝑍𝑍𝜀𝜀𝜀𝜀𝜀2∕𝑆𝑆2 | ; it is
|𝜀𝜀𝑀𝑀2 |
𝐴𝐴 to 𝐴𝐴 = |𝑍𝑍𝜀𝜀𝜀𝜀𝜀2∕𝑆𝑆2 | [𝑃𝑃 𝑃𝑃∕𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛], because

proportional 𝐴𝐴 𝐴𝐴𝐴 𝐴𝐴 and earth tidal strain amplitude
𝐴𝐴 |𝜀𝜀𝑀𝑀2 | are constant. The
earth tidal signal, as expected given the amplitude spectrum (Figure 3d), is small and explains less than 1 mm
of head oscillation in the borehole over a total oscillation of around 2 cm (Figure 3a), whereas approximately
1.9 cm is related to atmospheric loading. This relative faintness for earth tide explains why the phase lag signal
(Figure 4b) is much noisier, even if a global increasing trend from a mean of 37° before 2013 and a mean of 45°
between 2013 and 2018 can be noticed. There is no clear trend in the earth tide amplitude from 2008 to 2018.
The existing models that predict a positive phase lag for earth tide all require a source term due to vertical flow,
either in the form of pressure diffusion to the water table or leakage to another layer, like in (Rojstaczer, 1988;
Wang et al., 2018). They both hypothesize a constant head boundary condition on top, which is consistent in
the presence of a water table aquifer. Yet, in these two cases the barometric
phase lag observed in an open well should be the same as the earth tide phase
lag—considering the minus sign on the definition of the barometric phase,
( )
𝐴𝐴 is, arg −𝑝𝑝 𝑤𝑤 . Because we observe a negative phase lag for barometric
that
𝜌𝜌𝜌𝜌𝜌
𝑎𝑎
tide and a positive phase lag in the earth tide, we will derive in the next
section a new model adapted to our geological observations (Figure 2). This
model will also differ by the fact that it makes no further assumptions on
layer parameters, like equal loading efficiencies between the two layers done
in Rojstaczer (1988), or negligible storage in the top layer done in (Wang
et al., 2018). The derivation of the response (Section 4) will thus be more
general and the inversion of the model adaptable to other sites, even if it will
obviously require adapted justified assumptions.
4. Tidal Response of a Confined Bi-Layer Aquifer

4.1. Conceptual Model of a Confined Bi-Layer Aquifer
We here derive the response of a well𝐴𝐴level 𝐴𝑤𝑤 (𝑚𝑚) opened to a confined
bi-layer aquifer to barometric loading 𝐴𝐴 𝐴𝐴𝑎𝑎 (𝑃𝑃𝑃𝑃) and earth tidal strain
𝐴𝐴 𝐴𝐴 (𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛). The derivation is presented in the case of barometric load-
ing, but the differences between the two sources, which lay in the bound-
ary conditions, are presented, as well as the final solution. The aquifer is
supposed infinite in the horizontal direction, homogeneous and isotropic.
Following the adaptation of the Hantush and Jacob (1955)'s model to tidal
strain by Wang et al. (2018), the well is opened to the “bottom layer” of
the aquifer where flow is supposed horizontal, thus radial by symmetry. In
Fond Lahaye, the bottom layer (Figure 5) corresponds to the fractured ande-
site (Figure 2). This layer will drain a second layer (leakage), referred to as
Figure 5. Conceptual model of a confined bi-layer aquifer. A bilayer aquifer
is submitted to periodic sources of defined frequencies: earth tidal strain and the “top layer”, which corresponds in Fond Lahaye to the altered andesite
barometric pressure. The water level in a well opened to this aquifer responds (Figure 2). As in Wang et al. (2018), flow is assumed vertical in this layer,
to the sources. The phase and amplitude of the response is a function of but contrary to it, storativity of the top layer is non-zero, thus vertical flow
hydrodynamic (transmissivity T and storativity S of the bottom layer, vertical is diffusive. As the aquifer is confined by the clay matrix (Figure 2), the
hydraulic conductivity K’ and storativity S’ of the top layer), poroelastic
boundary conditions will be no-flow at the top and bottom confining layers
(Loading efficiency 𝐴𝐴 𝐴𝐴 & 𝛾𝛾 ’, Skempton coefficients time undrained bulk moduli
𝐴𝐴 𝐴𝐴𝐴𝐴𝑢𝑢 & 𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′ ) as well as geometrical parameters (layer width b & b', well (Figure 5). The hydraulic𝐴𝐴head 𝐴 (𝑚𝑚) in the bottom layer follows the diffusion
and casing 𝐴𝐴 radii 𝐴𝐴𝑤𝑤 & 𝑟𝑟𝑐𝑐 ). The radial axis r and the vertical z correspond to equation:
the reference frame used to describe the boundary conditions. Adapted from ( )
McMillan et al. (2019). 𝜕𝜕𝜕 𝛾𝛾 𝜕𝜕𝜕𝜕𝑎𝑎
(9) 𝑇𝑇 ∇2 ℎ + 𝑞𝑞 = 𝑆𝑆 −
𝜕𝜕𝜕𝜕 𝜌𝜌𝜌𝜌 𝜕𝜕𝜕𝜕
where 𝐴 is the hydraulic head in the bottom𝐴𝐴layer, 𝐴𝐴 the hydrogeological stor-

𝐴𝐴
age coefficient
𝐴𝐴 𝐴𝐴 (−), 𝐴𝐴 the transmissivity of the bottom𝐴𝐴layer 𝑚𝑚 𝐴𝐴 ⋅ s−2 , 𝐴𝐴 the
( )
loading efficiency of the bottom𝐴𝐴layer (−)

𝐴𝐴, and 𝐴𝐴 𝑚𝑚 ⋅ 𝑠𝑠−1 is the vertical leakage from the top layer. The boundary
( )
conditions are:
(10)
ℎ(𝑟𝑟𝑟 𝑟𝑟) = ℎ∞ (𝑡𝑡) 𝑎𝑎𝑎𝑎𝑎𝑎 = ∞,
𝑃𝑃
ℎ(𝑟𝑟𝑟 𝑟𝑟) = ℎ𝑤𝑤 (𝑡𝑡) + 𝑎𝑎 𝑎𝑎𝑎𝑎𝑎𝑎 = 𝑟𝑟𝑤𝑤 ,
(11)
𝜌𝜌𝜌𝜌
𝜕𝜕𝜕 || 𝜕𝜕𝜕
(12)
2𝜋𝜋𝜋𝜋𝑤𝑤 𝑇𝑇 = 𝜋𝜋𝜋𝜋2𝑐𝑐 𝑤𝑤 ,
𝜕𝜕𝜕𝜕 ||𝑟𝑟=𝑟𝑟𝑤𝑤 𝜕𝜕𝜕𝜕
where 𝐴𝐴𝑐𝑐 is the well casing inner radius

𝐴𝐴 𝐴𝐴 and 𝐴𝐴𝑤𝑤 the radius of the screened portion of the well (Figure 5). Note that
we here solve for the response to barometric loading, but the same can be applied to earth tidal strain by substi-
tuting in Equation (9) the source𝐴𝐴term 𝐴𝐴𝐴𝐴𝐴𝐴 to 𝐴𝐴 𝜕𝜕𝜕𝜕𝑎𝑎 , 𝐴𝐴
where 𝐴𝐴 is the Skempton coefficient
𝐴𝐴 and 𝐴𝐴𝑢𝑢 the undrained
𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
𝑢𝑢
𝜕𝜕𝜕𝜕
bulk modulus of the bottom layer. Boundary conditions are kept the same except for Equation (11), which applies
to an open well under barometric loading. For a closed well, a well equipped with a
packer, or for earth tides, Equation (11) simplifies to
ℎ(𝑟𝑟𝑤𝑤 , 𝑡𝑡) = ℎ𝑤𝑤 (𝑡𝑡)
4.2. Top Layer Response: Computation of the Vertical Leakage

𝑨𝑨 𝑨𝑨
Firstly, to get the expression 𝐴𝐴 of 𝐴𝐴 we must solve for the excess pore pressure in the
top𝐴𝐴layer, 𝐴𝐴′ (𝑃𝑃 𝑃𝑃). Since we assumed vertical flow, it follows a 1-D diffusion equation
with a source term:
( ′ )
𝜕𝜕 2 𝑝𝑝′ 𝑆𝑆 ′ 𝜕𝜕𝜕𝜕 ′ 𝜕𝜕𝜕𝜕𝑎𝑎
(13) = − 𝛾𝛾
𝜕𝜕𝜕𝜕2 𝐾𝐾 ′ 𝑏𝑏′ 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
where
𝐴𝐴 𝐴𝐴 ′ is the hydrogeological storage coefficient
𝐴𝐴 𝐴𝐴(−), 𝐴𝐴 ′ the vertical hydraulic
conductivity of the top𝐴𝐴layer 𝑚𝑚 𝐴𝐴
⋅ 𝑠𝑠 , 𝐴𝐴 (𝑚𝑚) the width of the top layer 𝐴𝐴 and 𝐴𝐴 ′ (−) the
( −1
) ′
loading efficiency of the top layer (e.g., Wang, 2000). We are looking for periodic
solutions at an angular frequency
𝐴𝐴 𝐴𝐴, thus physical quantity
𝐴𝐴 𝐴𝐴 will be expressed in the
frequency domain by its complex amplitude labeled with a symbol 𝐴𝐴 𝐴𝐴̃ . The boundary
Figure 6. Red lines: ( Chart of)the transfer ( ̃ function
) (eq 19 or eq 22) in terms conditions are, no flow on top:
𝜌𝜌𝜌𝜌 ℎ̃
of phase
𝐴𝐴 lag 𝐴𝐴𝑛𝑛 = 𝑎𝑎𝑎𝑎𝑎𝑎 (𝛾𝛾−1)𝑤𝑤𝑃𝑃̃ 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 𝐵𝐵𝐵𝐵 𝑤𝑤𝜀𝜀̃ and adimensional (normalized)
𝑎𝑎 𝑢𝑢
𝜌𝜌𝜌𝜌 ℎ̃ 𝜌𝜌𝜌𝜌 ℎ̃ 𝜕𝜕 𝑝𝑝̃′
gain 𝐴𝐴𝑛𝑛 = | (𝛾𝛾−1)𝑤𝑤𝑃𝑃̃ | 𝑜𝑜𝑜𝑜| 𝐵𝐵𝐵𝐵 𝑤𝑤𝜀𝜀̃ |, as a function of the borehole storage parameter (14) =0
𝑎𝑎 𝑢𝑢 𝜕𝜕𝜕𝜕 |𝑧𝑧=𝑏𝑏′
𝐴𝐴 i.e., 𝐴𝐴 = +∞). Green lines:
𝐴𝐴 (In this abacus we assume no leakage effect,
Chart drown from Hsieh et al. (1987). And pore pressure continuity at the layer interface:
(15)
𝑝𝑝̃′ (𝑧𝑧 = 0) = 𝜌𝜌𝜌𝜌 ℎ̃
𝐴𝐴with ℎ̃ the complex amplitude of the hydraulic head in the bottom layer. The solution to this equation is (see
Appendix A):
( (√ ) (√ ) (√ ))
𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 ′ 𝑖𝑖𝑖𝑖
(16) ̃ ′ ̃
+ 𝛾𝛾 ′ 𝑃𝑃̃𝑎𝑎
( )
̃′
𝑝𝑝 (𝑧𝑧) = 𝜌𝜌𝜌𝜌 ℎ − 𝛾𝛾 𝑃𝑃𝑎𝑎 cosh 𝑧𝑧 − tanh 𝑏𝑏 sinh 𝑧𝑧
𝐷𝐷′ 𝐷𝐷′ 𝐷𝐷′
𝐾𝐾 ′ 𝑏𝑏′
𝐴𝐴with 𝐴𝐴′ = the hydraulic diffusivity of the top𝐴𝐴layer 𝑚𝑚2 ⋅𝐴𝐴𝑠𝑠−1 , 𝐴𝐴 water density (𝐴𝐴𝐴𝐴 ⋅ 𝑚𝑚−3𝐴𝐴
) and 𝐴𝐴 grav-
( )
𝑆𝑆 ′
𝐴𝐴
′2
ity acceleration
𝐴𝐴 (𝐴𝐴 ⋅ 𝑠𝑠−2). We see in this intermediate result the dimensionless parameter 𝐴𝐴 𝐴𝐴 = 𝑏𝑏2𝐷𝐷𝜔𝜔′ defined by
Rojstaczer (1988). Thus, the source𝐴𝐴term 𝐴𝐴𝐴 in the bottom layer will be:
𝐾𝐾 ′ 𝜕𝜕 𝑝𝑝̃′
( )
𝐾𝐾 ′ ̃ 𝛾𝛾 ′ ̃
(17)
√ √
𝑞𝑞̃ = =− ′ ℎ− 𝑃𝑃𝑎𝑎 (1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄)
𝜌𝜌𝜌𝜌 𝜕𝜕𝜕𝜕 |𝑧𝑧=0 𝑏𝑏 𝜌𝜌𝜌𝜌
4.3. Whole Aquifer Response: Computation of the Head in the𝑨𝑨Well 𝒉𝒉

̃ 𝒘𝒘
Now we may solve the hydraulic head in the bottom layer. Assuming radial symmetry, and injecting
𝐴𝐴 𝐴𝐴 in (9),
(17)
𝐴𝐴 ̃ℎ follows:
( 2 ) ( ) ( )
𝜕𝜕 ℎ̃ 1 𝜕𝜕 ℎ̃ 𝐾𝐾 ′ ̃ 𝛾𝛾 ′ ̃ √ √
̃ − 𝛾𝛾 𝑃𝑃̃𝑎𝑎
(18)
𝑇𝑇 + − ℎ − 𝑃𝑃 𝑎𝑎 (1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄) = 𝑖𝑖𝑖𝑖𝑖𝑖 ℎ
𝜕𝜕𝜕𝜕2 𝑟𝑟 𝜕𝜕𝜕𝜕 𝑏𝑏′ 𝜌𝜌𝜌𝜌 𝜌𝜌𝜌𝜌
From here the solving process is the same as in Wang et al. (2018), thus it is left in Appendix B. The general
solution we obtain is:
√
⎛ 𝑆 (1 + 𝑖) 𝑄 ⎞
⎜ (1 − 𝛾') + (1 − 𝛾) ⋅ ′
( √ )⎟
𝑆 tanh (1 + 𝑖) 𝑄 ⎟
𝜌𝑔̃ℎ𝑤 ⎜⎜ ⎟⋅ 1
(19) = √
̃
−𝑃𝑎 ⎜ (1 + 𝑖) 𝑄 ⎟ 𝜉
⎜ 𝑆 ⎟
1+ ′ √
⎜ 𝑆 tanh((1 + 𝑖) 𝑄) ⎟
⎝ ⎠
Table 1 where
Phase Lag Sign, in the Absence of Borehole Storage, Depending on the
Relative Values of Corresponding Poroelastic Parameters
𝑊𝑊 𝐾𝐾0 (𝛽𝛽𝛽𝛽𝑤𝑤 )
(20)
𝜉𝜉 = 1 + 𝑖𝑖 ⋅ 𝑐𝑐𝑟𝑟
𝛽𝛽𝛽𝛽𝑤𝑤 𝐾𝐾1 (𝛽𝛽𝛽𝛽𝑤𝑤 )
Barometric tide Earth tide
)1
𝐴𝐴 𝐴𝐴 𝐴 𝐴𝐴 ′
𝐴𝐴 𝐴𝐴 𝐴 𝐴𝐴 ′
𝐴𝐴 𝐴𝐴𝐴𝐴𝑢𝑢 < (𝐵𝐵𝐵𝐵𝑢𝑢 )′𝐴𝐴 𝐴𝐴𝐴𝐴𝑢𝑢 > (𝐵𝐵𝐵𝐵𝑢𝑢 )′
( √
(1 + 𝑖𝑖)tanh((1 + 𝑖𝑖) 𝑄𝑄) ′ 2
(21)
√
𝐴𝐴 + 𝐴𝐴 − 𝐴𝐴 − 𝐴𝐴 + 𝛽𝛽𝛽𝛽𝑤𝑤 = 𝑊𝑊 √ 𝑆𝑆 + 2𝑖𝑖𝑖𝑖
𝑄𝑄
where Ko and K1 are the modified Bessel functions of the second kind,
𝑟𝑟2𝑤𝑤 𝜔𝜔
respectively of the zeroth and first order
𝐴𝐴 and 𝐴𝐴 = 2𝑇𝑇
. In the case of Earth
tide response, the solution, detailed in the appendix, is:
√
⎛ ′ ′ 𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄 ⎞
⎜ 𝐵𝐵 𝐾𝐾𝑢𝑢 + 𝐵𝐵𝐵𝐵𝑢𝑢 ⋅ ′
𝑆𝑆 tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎟ 1
√ ⎟
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤 ⎜
(22) =⎜ ⎟ ⋅ 𝜉𝜉
𝜀𝜀̃
√
⎜ 𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄 ⎟
1 + ′
𝑆𝑆
⎜ √ ⎟
⎝ tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎠
To study both responses in parallel, it is useful to note that substituting the source
𝐴𝐴 𝐴𝐴𝑃𝑃̃𝑎𝑎 by 𝐴𝐴𝐴 can be achieved in the
response by substituting the poroelastic parameters
𝐴𝐴 (𝛾𝛾 −𝐴𝐴1) by 𝐴𝐴𝐴𝐴
𝐴𝐴𝑢𝑢 and (𝛾𝛾 −𝐴𝐴1) by 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′. That said, both responses
′
are controlled by the same four adimensional parameters:

)2
(
𝑟𝑟𝑐𝑐 𝑏𝑏′2 𝜔𝜔 𝑆𝑆
𝑊𝑊 = ; 𝑐𝑐𝑟𝑟 = ; 𝑄𝑄 = and ′ .
2𝑇𝑇 𝑟𝑟𝑤𝑤 2𝐷𝐷′ 𝑆𝑆
4.4. Physical Meaning of the Four Adimensional Parameters

Firstly,
𝐴𝐴 𝐴𝐴𝐴𝐴 and 𝐴𝐴𝑟𝑟 were defined in Hsieh et al. (1987) in the case of earth tide and dictate the borehole storage
effect, that is, a concurrent negative phase lag and attenuation. As shown on Figure 6, our model𝐴𝐴with 𝐴𝐴′ = 0 is
in perfect agreement with the model of a confined aquifer of Hsieh et al. (1987). It is important to note that in
this case, the phase lag and gain for barometric pressure and earth tide are the same. A 𝐴𝐴 low 𝐴𝐴 indicates a low
borehole storage effect, which vanishes as soon 𝐴𝐴 as 𝐴𝐴 𝐴𝐴𝑟𝑟 < 10−2 (Figure 6 and Hsieh et al., 1987), while 𝐴𝐴
a high 𝐴𝐴
indicates a strong borehole storage effect.
𝐴𝐴 𝐴𝐴𝑟𝑟 modulates this effect given the geometry of the borehole (Figure 6;
( )2
Table 1) and equals in our case 𝐴𝐴 to 𝐴𝐴𝑟𝑟 = 𝑟𝑟 𝑐𝑐 = 0.42.
𝑟𝑟
𝑤𝑤
A second effect adds up to this first: the leakage effect, which is controlled 𝐴𝐴 by𝐴𝐴𝐴𝐴 and 𝑆𝑆𝑆𝑆′ . This effect is illustrated in
the case of phase lag and gain for barometric pressure on Figures 7c and 7d, respectively and in the case of phase
lag and gain for earth tide on Figures 7a and 7b, respectively. In Figure 7, we assume no borehole storage effect,
that
𝐴𝐴 is, 𝐴𝐴 𝐴𝐴𝑟𝑟 < 10−2. The interpretation is less straightforward since now there are 3 distinct regimes (as identified
in the case of barometric loading by Rojstaczer (1988)).
• A high frequency regime ( 𝑄𝑄 𝑆𝑆𝑆𝑆′ ≫ 1, right of sub-figures 7) where the diffusion is negligible and the flow
√
𝐴𝐴
is controlled by the bottom layer. Phase lag tends 𝐴𝐴 to 0◦ in both earth tide (Figure 7a) and barometric loading
(Figure 7c). According to Figure 7d), the barometric gain is equal 𝐴𝐴 to (1 − 𝛾𝛾) (bottom layer loading efficiency)
and the earth tide gain is equal
𝐴𝐴 to BKu. We assumed this regime when studying independently the borehole
storage effect on Figure 6.
• A low frequency regime (on the left of sub-Figure 7,𝐴𝐴 when 𝑄𝑄 𝑆𝑆𝑆𝑆′ ≪ 1), where vertical diffusion is strong
√
enough so that the two layers are coupled: phase𝐴𝐴lag is 0◦ in both earth tide and barometric loading (Figures 7a
and 7c) and the amplitude is controlled by the poroelastic coefficient of the layer with highest storage. More
𝑆𝑆 ′ (1−𝛾𝛾 ′ ) + 𝑆𝑆(1−𝛾𝛾)
precisely it equals to the weighted𝐴𝐴mean 𝑆𝑆 ′ + 𝑆𝑆
for barometric gain and tends 𝐴𝐴 to 1 − 𝛾𝛾 ′𝐴𝐴when 𝐴𝐴 𝐴 𝐴𝐴 ′
𝑆𝑆 ′ 𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′ + 𝑆𝑆𝑆𝑆𝑆𝑆𝑢𝑢
(Figure 7d) and equals
𝐴𝐴 to 𝑆𝑆 ′ + 𝑆𝑆
and𝐴𝐴tends 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′𝐴𝐴when 𝐴𝐴 𝐴 𝐴𝐴 ′ to for earth tide gain (Figure 7b).
( ̃ )
𝜌𝜌𝜌𝜌 ℎ̃
Figure 7. Chart of the transfer functions for earth tide (eq 22): (a) Phase 𝐴𝐴 lag 𝐴𝐴 = 𝑎𝑎𝑎𝑎𝑎𝑎 𝜀𝜀̃ 𝑤𝑤 and (b)
𝐴𝐴 gain 𝐴𝐴 = | 𝜀𝜀̃ 𝑤𝑤 |. For barometric pressure (Equation 19): (c) Phase
( ̃ ) ̃
𝐴𝐴 lag 𝐴𝐴 = 𝑎𝑎𝑎𝑎𝑎𝑎 −𝑃𝑃̃𝑤𝑤 and (d)
𝐴𝐴 gain 𝐴𝐴 = | −𝑃𝑃̃𝑤𝑤 |, as a function of the adimensional parameter Q (In this chart we assume no borehole storage, 𝐴𝐴 i.e., 𝐴𝐴 𝐴𝐴𝑐𝑐 < 10−2).
𝜌𝜌𝜌𝜌 ℎ 𝜌𝜌𝜌𝜌 ℎ
𝑎𝑎 𝑎𝑎
n intermediate frequency regime

A ( 𝑄𝑄 𝑆𝑆𝑆𝑆′ ∼ 1) where phase shift occurs thanks to the exchange between
√
• 𝐴𝐴
the layers. The phase shift sign depends on the relative values of the poroelastic parameters (Table 1). The
phase lag maximum or minimum values attainable are controlled by the contrast between these parameters,
(higher contrast induces larger phase lag). In this regime, amplitude smoothly varies between the two previous
extreme amplitudes.
4.5. Model Constraint
If the model encompasses many possible behaviors, we will show that looking at the phase shifts related
to earth and barometric tides enables to constrain the model down to a few varying parameters that can be
inverted.
Firstly, earth tide phase lag gives out a lot of information. Figure 8a shows the maximum attainable earth tide
phase lag for all value
𝐴𝐴 of 𝐴𝐴𝐴 𝐴𝐴𝐴𝐴𝑢𝑢 ,𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′ as a function of W and S/S’. The mean earth tide phase lag in the Fond
Lahaye borehole varies between 37° (before 01/2013) and 45° (after 01/2013) (red dotted lines on Figures 8a
and 8b). Thus, from Figure 8a, we can infer that the borehole storage effect must be negligible at least start-
ing from 2013 (since borehole storage only decreases the phase lag), and
𝐴𝐴 that 𝐴𝐴 = 2𝑇𝑇
≤ 10−2. The remaining
question, is whether borehole storage (change in W) or vertical diffusion (change in Q) is responsible for the
change in phase lag between these two intervals. A change in parameter W implies that phase lag increases
simultaneously and with the same amount for barometric pressure and earth tidal signals (Figure 6). Yet, the clear
increasing trend observed in barometric phase lag in 2009/2010 (Figure 4a) is not seen on earth tide (Figure 4b),
which increase occurred between 2011 and 2013. Because the changes in phase lag are not simultaneous, we
( )
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤
Figure 8. (a) Maximum attainable earth tide phase
𝐴𝐴 lag 𝐴𝐴 = 𝑎𝑎𝑎𝑎𝑎𝑎 𝜀𝜀̃
for all values
𝐴𝐴 𝐴𝐴of 𝐴𝐴, 𝐴𝐴𝐴𝐴
𝐴𝐴 𝑢𝑢 and 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′, depending
𝐴𝐴 on 𝐴𝐴
𝐴𝐴 and 𝑆𝑆
𝑆𝑆 ′
. Maximum values are obtained 𝐴𝐴
setting 𝐴𝐴 ′
𝐾𝐾𝑢𝑢′ = 0 (b) Chart of earth tide phase
𝐴𝐴 lag 𝐴𝐴 predicted by the model for different
𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′
values 𝐴𝐴 ′ and
𝐴𝐴 of 𝐴𝐴∕𝑆𝑆 𝐵𝐵𝐵𝐵𝑢𝑢
, with neglectable borehole storage 𝐴𝐴
effect (𝐴𝐴 𝐴 1). The maximum of the curve
𝐴𝐴 with 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′ = 0 (label
‘+’) on (b) corresponds to a dot for a given S/S’ on (a) for W = 10 −3. a & b, red dotted lines: mean observed earth tide phase
lag before and after 01/2013.
conclude that borehole storage (adimensional parameter W) is negligible for the whole studied period and set
𝐴𝐴 𝐴𝐴 = 2𝑇𝑇
≤ 10−2, which translates in our response functions Equations (19 & 22), in:
(H1)
𝜉𝜉 = 1.
𝐴𝐴 of 𝐴𝐴𝐴 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′ ∕𝐵𝐵𝐵𝐵𝑢𝑢 and S/S’ with the hypothesis

Figure 8b, shows the evolution of earth tide for different values
𝐴𝐴 𝐴𝐴 ≤ 10−2. Our data range between 37° (before 01/2013) and 45° (after 01/2013) (red dotted lines). The positive
sign of the phase lag tells us
𝐴𝐴 that 𝐴𝐴𝐴𝐴𝑢𝑢 > 𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′ (Table 1). In addition, the range of the observed phase lag induced
a constrain on the𝐴𝐴ratio 𝐴𝐴 𝐾𝐾𝑢𝑢 ∕𝐵𝐵𝐵𝐵𝑢𝑢 (Figure 8b) and the phase lags only be predicted if:
′ ′
(H2) < 10−2 .
𝐵𝐵𝐵𝐵𝑢𝑢
From Gassmann's equation (Gassman, 1951), we know that

𝛼𝛼
𝐵𝐵𝐵𝐵𝑢𝑢 = 𝜙𝜙 ,
(H3) + 𝛼𝛼−𝜙𝜙
𝐾𝐾 𝐾𝐾 𝑓𝑓 𝑠𝑠
where
𝐴𝐴 𝐴𝐴 is the Biot coefficient,
𝐴𝐴 𝐴𝐴 is the porosity,
𝐴𝐴 𝐴𝐴𝑓𝑓 the bulk modulus of
the fluid
𝐴𝐴 and 𝐴𝐴𝑠𝑠 the bulk modulus of the solid grain. The hypothesis
𝐴𝐴 (𝐻𝐻2)
implies that the top layer is more porous that the bottom layer and that grain
bulks modulus of the top layer is lower than the one of the bottom layer. This
is consistent with the geological observations (Figure 2) as the bottom layer
is a fractured andesite (thus low porosity) and the top layer is made of altered
andesite (higher porosity and lower grain bulk modulus).
Furthermore, when looking at Figure 8b, we can observe that our data are
consistent𝐴𝐴with 𝑆𝑆𝑆𝑆′ ≤ 0.01, since only dark green and blue curves cross the
range of observed data. Assuming constant storativities is common prac-
tice when dealing with variable hydrodynamic parameters, because phase
Figure 9. Colored lines: Phase/Amplitude diagram of the √ theoretical
barometric response
𝐴𝐴 𝐴𝐴 with 𝐴𝐴 = 1 (𝐻𝐻1), and varying
(𝑒𝑒𝑒𝑒 19), 𝐴𝐴 𝑄𝑄 𝑆𝑆𝑆𝑆′ .The value of
responses are less sensitive to them (Hsieh et al., 1987; Rojstaczer, 1988;
𝐴𝐴 𝐴𝐴 and the
𝐴𝐴 ratio 𝐴𝐴 ′ ∕𝛾𝛾 is specified for each line. Green dots: observed barometric Wang et al., 2018). Thus, we chose to that the 𝐴𝐴ratio 𝑆𝑆𝑆𝑆′ is constant
response points in the Phase/Amplitude space. (Hypothesis H3).
Besides, barometric response function also yields information. It is represented in a Phase/Gain diagram
on Figure 9 for different value 𝐴𝐴 of 𝐴𝐴 (each color means a fixed value 𝐴𝐴 of 𝐴𝐴)
𝐴𝐴 and 𝐴𝐴 ′ ∕𝛾𝛾 . We already know that
𝐴𝐴 𝐴𝐴 ′ < 𝛾𝛾 given the negative phase lag (table 1), thus we explore the possible range of these poroelastic param-
eters. Our data appears in green in Figure 9. It appears that in the vicinity of data points, the model is quite
insensitive𝐴𝐴 to 𝐴𝐴 (the lines with different colors are close together), but far more sensitive 𝐴𝐴 to 𝐴𝐴 ′ (the lines with
differing markers are spaced), which makes 𝐴𝐴 of 𝐴𝐴 a good candidate to be assumed constant. We observe on
Figure 9 that to explain the entirety of phase lag variation, we need𝐴𝐴to set 𝐴𝐴 ≥ 0.9. This is consistent with the
geology of the bottom layer: compliant fractures in rigid andesite explain why all of the exerted pressure
has an effect on pore pressure (Brajanovski et al., 2005; Gallagher et al., 2022). For the sake of simplicity,
we set
(H4)
𝛾𝛾 ∼ 1.
If we do observe that the model is sensitive to the𝐴𝐴ratio 𝐴𝐴 ′ ∕𝛾𝛾 , the ratio should be in the range 0.05–0.2 to explain
our data. The last 3 hypotheses (H2, H3, H4) seem unrelated. Yet poroelasticity theory (Wang, 2000) tells us that
𝐴𝐴 = 𝐵𝐵 3(1−𝜈𝜈𝑢𝑢 ) , 𝐴𝐴
where 𝐴𝐴𝑢𝑢 is the undrained Poisson ration,
𝐴𝐴 and 𝐴𝐴𝑠𝑠 = , 𝐴𝐴
where 𝐴𝐴𝑠𝑠 is the uniaxial specific storage, 𝐴𝐴 is
1+𝜈𝜈 𝛼𝛼
𝐴𝐴 𝐾𝐾𝑣𝑣 𝛾𝛾
𝐴𝐴
𝑢𝑢
𝐴𝐴 and 𝐴𝐴𝑣𝑣 the uniaxial drained bulk modulus. Thus, it is expected that:
Biot's coefficient
𝐵𝐵 ′ 𝛾𝛾 ′ 𝑆𝑆
(23)∝ ∝ ′,
𝐵𝐵 𝛾𝛾 𝑆𝑆
which is consistent with our case where

𝐴𝐴 𝐴𝐴 ′𝐴𝐴≪ 𝛾𝛾 , 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′ ≪ 𝐵𝐵𝐵𝐵𝐴𝐴
𝑢𝑢 and 𝐴𝐴 𝐴 𝐴𝐴 . In addition, the hypotheses are
′
consistent with the geological observation: a bottom layer made of fractured andesite (small porosity, but very
compressible) and a top layer made of altered andesite (higher porosity but less compressible).
4.6. Model Inversion

Thanks to the three assumptions
𝐴𝐴 (𝐴𝐴1 − 4), Equation (19) reduces to:
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤 𝛾𝛾 ′ − 1
=
̃
(24)
𝑃𝑃𝑎𝑎
√
(1 + 𝑖𝑖) 𝑄𝑄
𝑆𝑆
1+ ′
𝑆𝑆 tanh((1 + 𝑖𝑖) 𝑄𝑄)
√
This simplified model is able to cover the range −45 to 0° in terms of phase lag. We are left with three parameters:
𝐴𝐴 𝐴𝐴 𝐴𝐴 ′, 𝐴𝐴𝐴 S/S’. We assume S/S’ constant and we fix it at𝐴𝐴10 −2 (𝐻𝐻3). Note that a change in the𝐴𝐴ratio 𝐴𝐴∕𝑆𝑆’ will not the
change fundamentally the results, especially the relative change of the hydrodynamic properties with time. Since
we have two independent observations (Amplitude and phase lag) we can perform a numerical inversion of these
last two parameters (Note𝐴𝐴that if 𝑄𝑄 𝑄 10, the inversion can be analytical – see appendix C). It yields the results
√
presented on Figure 10.

Following the same process, and the three assumptions 𝐴𝐴 (𝐴𝐴1 − 3), Equation (22) reduces to:
(
)
𝑆𝑆 √
𝐵𝐵𝐵𝐵𝑢𝑢 ⋅ (1 + 𝑖𝑖) 𝑄𝑄 +
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤 𝑆𝑆 ′ 𝐵𝐵𝐵𝐵𝑢𝑢
(25) =
𝜀𝜀̃
√
𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄
1+ ′
𝑆𝑆 tanh((1 + 𝑖𝑖) 𝑄𝑄)
√
Which covers the range 0 to +45° in terms of phase lag.𝐴𝐴Since 𝑄𝑄 , we are left with two parameters:
𝑆𝑆
√
𝐵𝐵𝐵𝐵𝑢𝑢
≪ 𝑆𝑆 ′
𝐴𝐴 𝐴𝐴𝑢𝑢 and 𝐴𝐴. Again, a numerical inversion was performed, yielding the results presented in Figure 11.
𝐴𝐴𝐴𝐴
Figure 10a shows the evolution of the diffusivity D’-deduced from the adimensional parameter Q-in the top layer
inferred from the barometric loading analysis (Equation 24). From 2008 to 2018, the diffusivity increases by
more than one order of magnitude. In 2008, the diffusivity is found𝐴𝐴to be 2.5 10−5 𝑚𝑚2 ∕s; it increases to a value
𝐴𝐴 of 10−4 𝑚𝑚2 ∕s between 2010 and 2013, and increases again to a value
𝐴𝐴 of 5 ⋅ 10−4 𝑚𝑚2 ∕s between 2015 and 2018. In
addition to the long-term evolution, several temporary changes: several increases or decreases of D’ that can
Figure 10. Evolution of (a) top layer vertical Diffusivity

𝐴𝐴 𝐴𝐴′ and (b) top layer loading efficiency
𝐴𝐴 𝐴𝐴 ′ inverted from the
barometric transfer function. The orange circles correspond to the results of pumping tests presented in (4.7). Vertical brown
lines correspond to earthquakes, vertical blue lines correspond to heavy rainfall events and the orange/green rectangles
to the pumpings for drinking water operation of the aquifer. Uncertainty on these parameters, which mainly comes from
measurement uncertainty, are plotted in Figure S4 in Supporting Information S1.
be observed are transients. Such transient change of properties can also be observed on the loading efficiency
(Figure 10b).
The evolution of the diffusivity can also be obtained independently from the earth tide analysis (Equation 25).
Here we repeat that the piezometric oscillation amplitude are controlled by approximately 90% of the barometric
loading and 10% of the earth tide loading. Even if the results (Figure 11a) are noisy and yield more uncertainty,
the general tendency of the diffusivity D' is consistent with the previous one, with an average around
𝐴𝐴 5 ⋅ 10−5 𝑚𝑚2 ∕s
in 2008–2010 𝐴𝐴 and 5 ⋅ 10 𝑚𝑚 ∕s between 2013 and 2018. Yet it is far less interesting to study because it does not
−4 2
enable to identify transient behavior, and thus identify causes behind the observed evolutions. Finally, the evolu-
tion
𝐴𝐴 of 𝐴𝐴𝐴𝐴𝑢𝑢 is given in Figure 10b, with a mean value of 8 GPa, which seems to be a reasonable value (Bailly
et al., 2019; Larochelle et al., 2022; Wang et al., 2018).
4.7. Validation With Pumping Tests
The key point is: can we be confident with the evolution of the diffusivity found by the tidal analysis (Figure 10a)?
It turns out, that four pumping tests were done in this aquifer (Table 2). The characteristics and results of each
pumping test are listed in Table 2. They were obtained with the MLU software which can reproduce the same
geometry as described in Figure 5, with an analytical model similar to ours but adapted to pumping tests
(Maas, 1987). MLU parameter optimizer was used to obtain the parameters (hydraulic conductivity and storage)
yielding the best fit to the measured data. Measured data and fitted curves are presented in Figure S1 in Support-
ing Information S1, along with additional detail on MLU parameters. We will use these results to validate the
model.
Figure 11. Evolution of (a) top layer vertical Diffusivity

𝐴𝐴 𝐴𝐴′ inverted from earth tide transfer function
𝐴𝐴 (b) 𝐴𝐴𝐴𝐴𝑢𝑢 (𝐺𝐺𝐺𝐺 𝐺𝐺)
with B the bottom layer Skempton coefficient 𝐴𝐴 and 𝐴𝐴𝑢𝑢 the undrained bulk modulus. The orange circles correspond to
the results of pumping tests presented in (4.7). Vertical brown lines correspond to earthquakes, vertical blue lines
correspond to heavy rainfall events and the orange/green rectangles to the pumpings for drinking water operation of the
aquifer.
𝐴𝐴 is 2.4 ⋅ 10−4 𝑚𝑚∕s (Table 2)

First, the lowest measured transmissivity of the bottom layer obtained by pumping tests
yields 𝐴𝐴 𝐴𝐴𝑐𝑐 ∼ 10 . This first observation validates the hypothesis
𝐴𝐴 −5
𝐴𝐴 (𝐴𝐴1): the borehole storage effect is negligible
because bottom layer transmissivity is high enough.
The second observation is that despite being less sensitive
𝐴𝐴 to 𝐴𝐴∕𝑆𝑆 ′ than to transmissivity, all fitted curves of the
pumping test analysis by the MLU software were compatible with a constant storativities (Hypothesis H3), and
the pumping test analysis leads
𝐴𝐴 to 𝑆𝑆𝑆𝑆′ = 0.01.
Finally, the results of the model in terms of top layer diffusivity are validated by these pumping tests, as seen
on Figures 10a (and 11a). Both absolute values 𝐴𝐴 (from 𝐴𝐴′ = 10𝐴𝐴−5 to 10−4 𝑚𝑚2 ∕s) and increasing trend are repro-
duced, and this despite the large discrepancy between the different conditions in which the pumping tests were
Table 2
Pumping Tests Characteristics and Results in Terms of Hydrodynamic Parameters
Distance to Flow rate Aquifer transmissivity Top layer
( diffusivity Storativity
Year pumping well Duration (m 3/h) 𝐴𝐴 ratio 𝐴𝐴∕𝑆𝑆 ′ (−)
( ) )
𝐴𝐴 𝐴𝐴 𝑚𝑚2 ∕𝑠𝑠 𝐴𝐴 𝐴𝐴′ 𝑚𝑚2 ∕𝑠𝑠
2007 180 m 72 hr 17.2 𝐴𝐴 5.7 ⋅ 10−4 𝐴𝐴 2.2 ⋅ 10−5 0.01
2008 180 m 72 hr 25.0 𝐴𝐴 5.23 ⋅ 10−4 𝐴𝐴 3.5 ⋅ 10−5 0.01
2013 180 m 42 days 30.0 𝐴𝐴 3.4 ⋅ 10 −4
𝐴𝐴 3.5 ⋅ 10 −5 0.01
2022 0m 4.56 hr 3.5 𝐴𝐴 2.4 ⋅ 10 −4
𝐴𝐴 2.5 ⋅ 10 −4 –
Figure 12. Zoom of Figure 10 on the different earthquakes effect on diffusivity: (a) 2010, (b) 2015, and (c) 2017.
conducted, as their duration, flow rate and pumping well (with or without observation well) changed between
pumpings.
5. Discussion
5.1. Evolution of the Hydrodynamic Properties Over Time
5.1.1. Impact of Earthquakes
Literature has identified earthquakes as a major cause for changes in properties of aquifers (Rojstaczer et al., 1995;
C.-Y. Wang et al., 2004; Elkhoury et al., 2006; Manga et al., 2012; Vittecoq et al., 2020). In our case, it is possible
to link four earthquakes to change in diffusivity thanks to tidal analysis and an additional one thanks to pumping
tests. The two first are linked to transient increase: the 2007 for which pump-
ing tests indicated an increase in top layer diffusivity𝐴𝐴from 2.2 ⋅ 10−5 𝑚𝑚2 ∕s to
𝐴𝐴 3.5 ⋅ 10−5 𝑚𝑚2 ∕s (Figure 10, Table 2), and the 2010 one, for which diffusivity
directly increased𝐴𝐴from 4 ⋅ 10−5 𝑚𝑚𝐴𝐴
2
∕s to 8 ⋅ 10−5 𝑚𝑚2 ∕s, but came back to the
initial value in 3 months (Figure 12a). We attribute these transient effects
to unclogging of fractures, a phenomenon already reported both at the lab
and field scale in numerous studies (Barbosa et al., 2019; Boeut et al., 2020;
Candela, 2014; Elkhoury et al., 2011; Vittecoq et al., 2020; H. Wang
et al., 2020). Unclogging of fractures have been shown to be most plausible
phenomenon at stake in the intermediate field, where static strain change is
too low to have significant impact and dynamic strains are too low to cause
fracturation. Unclogging is able to explain the large variations in permeabil-
ity observed, especially in fractured aquifer. The 2014 earthquake has also
induced an increase in diffusivity. However, this earthquake occurred during
a long-term pumping test, thus we cannot conclude if increase of diffusivity
between 2013 and 2015 is related to the earthquake or to the pumping test.
It is, however, interesting to note that this earthquake has induced a clear
permanent increase of permeability in another aquifer of the island (Vittecoq
et al., 2020).
The next two identified earthquakes induced transient decreases in the top
layer diffusivity. In 2015 and 2017, diffusivity decreased𝐴𝐴from 5 ⋅ 10−4 𝑚𝑚2 ∕s
𝐴𝐴 and 3.5 ⋅ 10−4 𝑚𝑚2 ∕s before the earthquake
𝐴𝐴 to 2 ⋅ 10−4 𝑚𝑚2 ∕s after (Figures 12b
and 12c). Permeability decreases were already reported, namely in Shi
et al. (2019) or Vittecoq et al. (2020) which is attributed to clogging effect.
This decrease in permeability is transient, and after 2–3 months, the initial
Figure 13. Magnitude/distance plot of the earthquakes felt in Martinique value is recovered. This characteristic time of a few months can also be
between 2007 and 2018. Gray oblique lines correspond to constant seismic observed in the case of extreme rainfall events (Section 5.1.b), advocating for
energy density contours
𝐴𝐴 (𝐴𝐴 ∕𝑚𝑚3) (Wang & Manga, 2010) and dark brown one to a similar process in both case, particle mobilization. Given we only observed
a constant dynamic stress contour.
Figure 14. Zoom of Figure 10 on the different rainfall events effect on diffusivity: (a) 2008, (b) 2016 and (c) 2017.
three coseismic changes, it is difficult to be conclusive on a possible cause explaining why permeability would
increase or decrease after an earthquake. Previous hypotheses include the azimuth and the focal mechanism of the
earthquake, which is certainly relevant concerning near field effects, linked to static stresses, but less clearly in
the case of dynamic strains effects although it is probably relevant in fractured aquifer where preferential fracture
directions stand out (Shi et al., 2019; Xiang et al., 2019; Yan et al., 2014). If the underlying mechanism cannot be
demonstrated here, an interesting information can be inferred from the magnitude/distance plot of the earthquakes
that have an impact on the diffusivity (Figure 13). Roeloffs (1998), proposed a criterion for persistent water level
drops or𝐴𝐴rise: 𝐴𝐴 ≥ 1.88 log(𝑟𝑟) + 1.55, with M the magnitude and r the well-hypocenter distance in kilometers.
This empirical criterion was generalized by Wang and Manga (2010), based on seismic energy density, and
yields 𝐴𝐴 ≥ 0.7 log 𝑟𝑟3 + 𝐸𝐸 , 𝐴𝐴
where 𝐴𝐴 depends on the threshold of seismic energy density. These criterion were
( )
𝐴𝐴
designed to explain water level changes in wells, which are generally different from permeability enhancement,
as recognized in (Elkhoury et al., 2006; Wang & Manga, 2010). Here, such a criterion cannot explain why diffu-
sivity evolved after the 2010 earthquake while it did not after many others of similar energy (Figure 13). We thus
suspect that transient changes are better explained with a criteria on dynamic stress, which decreases as the square
of the distance (Aki & Richards, 2002). We graphically set our threshold𝐴𝐴to be 𝐴𝐴 ≥ 0.7 log 𝑟𝑟2 + 2.7 (Figure 13),
( )
by drawing the line of corresponding slope that was directly above all earthquakes impacting permeability. On
this figure, several earthquakes that were expected to have an impact could not be studied, either because they
were outside of the exploitable signal, or like in 2014, concurrent with pumpings in the aquifer which have a
strong impact on diffusivity. Trying to go further, we tried to relate the peak ground velocities (PGVs) to the
amplitude of change in the permeability. Using measured peak ground velocities (PGVs) as a proxy to dynamic
strain intensity, Elkhoury et al. (2006) observed a correlation between them.
Yet to the best of our knowledge, no similar observations was made since
then, as shown on Figure S3 in Supporting Information S1. This absence
of correlation of the data might be explained as the PGV and change in the
dynamic stress or strain is not straightforward.
5.1.2. Extreme Rainfall Events
A second phenomenon has been identified in a previous study in Marti-

nique: strong rainfall events (Vittecoq et al., 2020). Three similar transient
evolutions in the signal have been identified, in October 2008 July 2016
and May–August 2017. For these three events we observe a significant
decrease in the diffusivity, which decreased by a factor 2 to six over a
month, and a recovery over the same period (Figure 14). October 2008 was
reported to be the third rainiest October month since the start of rainfall
measurements in 1936. A total of 871 mm were recorded at the nearest
station (Météo France, 2008), with several days over 100 mm. July 2016
Figure 15. Zoom of Figure 10 on the consecutive pumpings in the aquifer in was also exceptionally rainy with almost 500 mm reported, concentrated
2013/2014. Orange rectangles refer to pumpings and green periods to recovery. on a few days, while the consecutive months of May–August 2017 have
The brown vertical lines represent the 2014 earthquake.
Table 3 known several events, from heavy rainfall to tropical storms, as already
Vertical Diffusion Length in the Top Layer, for the Tidal Model and the reported in (Vittecoq et al., 2020). Several mechanisms could explain the
Different Pumpings change in diffusivity. The first one is the fluidization of many solid parti-
Tidal model 2007 2008 2022 cles like colloids. Theory predicts that adding particles in a fluid increases
(2008–2018) pumping pumping 2013 pumping pumping its viscosity. In the case of rigid spherical particles with no interaction,
𝐴𝐴 𝐴𝐴 (m) 1.9–9.9 4.8 5.8 32.0 (whole layer) 3.0 in an incompressible fluid, the predicted change in viscosity is linked to
volume fraction of particles such that:
Note. The range depends mostly on the pumping duration and calculated
vertical diffusivity. 𝜇𝜇 1
=
𝜇𝜇0 1 − 25 𝜙𝜙
where 𝐴𝐴 is the dynamic viscosity of water with a volume fraction

𝐴𝐴 𝐴𝐴 𝐴𝐴 of particles,
𝐴𝐴 and 𝐴𝐴0 the dynamic viscosity of
water in the absence of particle (Kachanov & Abedian, 2015). This formula is in accordance with experimental
data up to a volume fraction of 30% when the non-interaction hypotheses fades. It could explain an increase up to
a factor 4 in the viscosity, thus a decrease in hydraulic conductivity and diffusivity by the same factor. This effect
fades once the colloids are evacuated, at the same rate of their appearance. A second explanation involves the
loading of the aquifer by temporary accumulation of rainfall water in the alluvium (Figure 2). Such an additional
load would increase the effective pressure in the aquifer, closing pathways for water to flow. Both explanations
would need further investigations to be tested, yet may contribute to the effect simultaneously. In both cases, the
observed delay of around 1 month for onset and decay of the decrease could be explained by pressure diffusion
and accumulation/drainage characteristic time.
5.1.3. Effect of Pumping in the Aquifer
We suspect a last phenomenon has had an impact on top layer diffusivity: long term pumping. It is difficult to
analyze the effect of pumpings with tidal analysis during the pumping because if the piezometric level decreases
too quickly, the tidal signal is lost. Yet we recovered many data points, especially during recovery periods, where
we manually checked the quality of the signal because quickly varying piezometric signals can affect tidal anal-
ysis. With all due precautions, the impact of pumpings on diffusivity, which varied by a factor 40 over the 2013
pumping and a factor 15 over the 2014, cannot be regarded as a random scatter, yet its effects are not evident
to analyze. Pumpings have a strong impact on the fluid velocity in the aquifer and on pore pressure. It seems
that diffusivity decreases during a pumping (see Figure 15 on July 2013), where the decline of D’ starts slightly
before the pumping because of the 29 days' time window. We consider two possible mechanism have a transient
effect during the pumping process: on the one hand particle mobilization in the fluid increases viscosity, that is,
decreases hydraulic conductivity. On the other hand, the increase in effective pressure
𝐴𝐴 𝐴𝐴 ∗ = 𝜎𝜎 − 𝑝𝑝 (here linked to
the decrease in pore pressure) closes pathways for water, that is, reduces fracture aperture and permeability. The
effect of recovery is more ambiguous, since during the first one diffusivity increased to recover its initial value.
The second recovery cannot be analyzed because of a possible impact of the earthquake, and during the third one,
diffusivity decreased. Further investigation would be needed to clearly conclude on the short-term impacts of
pumpings on diffusivity. About the long term impact, it seems that here the diffusivity increased overall, which
is understandable when considering that pumping can lead to unclogging, yet we cannot conclude as the 2014
earthquake was simultaneous and was shown to have had impacts on the hydraulic conductivity in another aquifer
of the island (Vittecoq et al., 2020).
5.2. Diffusivity Inferred From Pumping Tests and From Tidal Analysis
Sensitivity to parameters is different between a pumping test, which is only sensitive to flow phenomena, and a
tidal model, sensitive to poro-mechanical effects as well as flow ones. A pumping tests consists of a local pertur-
bation, which is mostly sensitive to transmissivity and vertical permeability in√ our model, on a horizontal extent
𝐴𝐴 𝐴𝐴 depending on the pumping duration
𝐴𝐴 𝐴𝐴𝑝𝑝 and bottom layer diffusivity
𝐴𝐴 𝐴𝐴 𝐴𝐴 : 𝐴𝐴 ∼ 𝐷𝐷𝐷𝐷𝑝𝑝 (Méité et al., 2022), and on
a vertical dimension , with 𝐴𝐴′ the vertical diffusivity in the top layer. Considering the tidal model,
√
𝐴𝐴 𝐴𝐴𝑝𝑝 ∼ 𝐷𝐷′ 𝑡𝑡𝑝𝑝𝐴𝐴
its sensitivity to parameters depend on the flow regime: in the absence of borehole storage effect, the head in the
aquifer is homogeneous and the response does not depend on√
transmissivity. The relevant space dimension in our
2𝜋𝜋𝜋𝜋′
case is the width of the diffusive process in the top𝐴𝐴layer 𝐴𝐴𝑡𝑡 ∼ 𝜔𝜔
. The different vertical diffusion lengths
𝐴𝐴 𝐴𝐴 are
summarized in table 3. Firstly, it appears
𝐴𝐴 that 𝐴𝐴𝑡𝑡 has evolved significantly during the decade 2008–2018 because
of the increase of vertical diffusivity. And since the width of the top layer is
14m, it corresponds effectively to new regions of the layer being reached.
Secondly, we can remark that the variety of the width investigated and pump-
ing durations, the results are consistent, which consolidates the homogeneity
assumption in our model.
5.3. Impact of Oceanic Tide
Oceanic tide is intuitively a good candidate given the proximity to the ocean
(2 km away). Yet several arguments show its contribution is negligible when
compared with that of earth tide. Firstly, it has been showed that Martinique
andesitic aquifers are compartmentalized (Vittecoq et al., 2015), such that
no diffusion of oceanic pressure is possible. This geological consideration is
striking when comparing different types of aquifers, all situated on the west
coast of Martinique. Those in highly transmissive pyroclastic flows made
of ashes are strongly coupled to the sea, as their spectrum show (Figure S2a
in Supporting Information S1), the peak at the K1 frequency is higher than
the one at the M2 frequency. On the contrary, all those in fractured andesite
show the same pattern in their spectrum (Figure S2b in Supporting Infor-
mation S1) with low or no peaks at the diurnal frequencies (O1, K1), and a
higher peak at the M2 frequency. It leaves a mechanical transmission of the
load the only possible process. Such a mechanical transmission would act
on the aquifer similarly as earth tide. Analyzing the remaining piezometric
signal after barometric influence is removed showed that oceanic tide signal
could not explain the observed amplitudes and its effect is at least an order of
magnitude lower than that of earth tide. Given that earth tide could already
be treated as a small correction to the barometric effect (the S2 frequency
is dominated by barometric contributions such that considering earth tide
only changes the barometric phase lag by around 2°), we conclude that our
Figure 16. Chart of normalized Earth tide response
𝐴𝐴 𝑤𝑤
plotted in the same
𝜌𝜌𝜌𝜌𝜌 results are unsensitive to oceanic loads and we neglect the correction linked
to oceanic loads which effect are below our measurement uncertainty.
𝐵𝐵𝐵𝐵 𝜀𝜀 𝑢𝑢
conditions as in (Sun et al., 2020, Figure
𝐴𝐴 1). 𝐴𝐴 = 10 , 𝑆𝑆 = 10−6 , 𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′ = 0.
′ −5
5.4. Comparison With Existing Models
Sun et al. (2020) proposed a review of the most important tidal response models, including (Hsieh et al., 1987;
Rojstaczer, 1988; Wang et al., 2018). Figure 16 presents the transfer function calculated in the same context. As
can be seen by comparing to Sun's charts, our model is closest to Rojstaczer (1988), who identified four imbal-
ances in his introduction, between Earth surface, water table, confining layer, the aquifer and the well. He focused
on the imbalance between the water table and the confining layer, neglecting the imbalance between the confining
layer and the aquifer by setting
𝐴𝐴 𝐴𝐴 = 𝛾𝛾 ′, (or equivalently
𝐴𝐴 𝐴𝐴𝐴𝐴𝑢𝑢 = 𝐵𝐵 ′ 𝐾𝐾𝑢𝑢′ , if he had dealt with earth tidal strain). As
already pointed in (Odling et al., 2015), his model was imprecise in the low frequency regime, that's why our
model predicts 0° phase lag𝐴𝐴when 𝐴𝐴 → 0, while Rojstaczer's tends to +45°. We did not consider the imbalance
at the water table because our two-layer aquifer is confined. On the contrary, we focus on the flow between the
two layers constituting the aquifer: the fractured andesite layer and the altered andesite layer (Figures 2 and 5).
Because of that, the most relevant model to compare to is from Brodsky and Prejean (2005), which studied the
exchange of pressure between a fault layer where permeability was assumed high (our main aquifer layer with
𝐴𝐴 𝐴𝐴 𝐴 1), and a host rock layer (similar to our aquitard layer). The main difference lies in the boundary condi-
tions, since they derived the model in the restrictive𝐴𝐴case 𝐴𝐴′ → +∞, in which our model coincide. In addition
to the geometry change, we coupled this generalized exchange model to the classical Hsieh et al. (1987) one to
take into account the borehole storage effect. Relatively to Rojstaczer's, the response changes significantly: the
positive phase lag induced by vertical leakage, and especially the low frequency behavior is totally different:
Rojstaczer (1988) predicts more attenuation with lower frequency, and a phase lag converging to −45°, while we
converge toward 0° of phase lag and a fixed value of attenuation, which depends on the poroelastic parameters
and the storativities. We also showed that the imbalance between the two layers can actually strongly influence
the response. Depending on the relative value of poroelastic parameters (Table 1) the flow direction changes and
Figure 17. Relative error on phase lag (left) and amplitude gain (right) between the responses at the two frequencies M2 and S2.
thus the sign of the response in absence of borehole storage. Its effect is particularly visible in our context where
storage is higher in the top layer, a more porous layer, than in the bottom layer made of fractured andesite.
Our model has a theoretical range of attainable phase lag between −135° and +45°, while previous model
(Rojstaczer, 1988; C.-Y. Wang et al., 2018) were limited to −90°. With the borehole storage effect, like previously,
we can reach the interval 0 to −90° (reaching −90° solely with borehole storage effect is actually unrealistic since
it requires that S→0, Hsieh's original model chart stopped around −80°). With the exchange between the layers,
we can reach −45° to +45° for both sources. Adding up all that we obtain the theoretical range of −135° to +45°.
Yet we should remember that with phase lags comes attenuation so it is unrealistic to observe good quality signals
with a phase lag of −135° (at least in the field case of Martinique where instruments have a mm resolution and
where shallow aquifers elastic coefficients (like bulk moduli) are relatively low. A more realistic range of obser-
vation of phase lags in confined aquifers therefore seems −100° to +45°.
5.5. Validity of Our Hypotheses and Uncertainty Assessment
A first hypothesis we made was that the response varied slowly with frequency, so that we could assume the
response was the same at M2 and S2 frequencies (1.93 and 2 cpd). In our range of parameters, the maximum
absolute difference in response induced by a similar frequency difference (3.5%) was inferior to 0.17° for
phase lag and 0.4% for gain, which is far below our measurement uncertainty, which is 1.9% for the gain
and 1.6° for phase lag (Duvall, 1927). The results in term of relative errors are plotted on Figure 17. Both
relative errors are below 3% on the full range of parameters. Another hypothesis we made was that the flow
in the top layer was vertical. The top layer of altered andesite in a clay matrix is opened to the well, contrary
to our conceptual model, in which the top layer was considered isolated from the well. Because the trans-
missivity of the top layer is so low compared to that of the bottom layer, the horizontal flow from that layer
is considered negligible with respect to the horizontal flow from the bottom layer. Yet even if it is far less
transmissive compared to the bottom layer, it is open to the well. What we implicitly did was neglecting the
horizontal flow coming from the top layer. To assess this hypothesis we can apply the (Hsieh et al., 1987)
model to the top layer and look at the resulting signal. For that we suppose that the horizontal diffusivity in
the top layer is the same as its vertical diffusivity and look at the worst case (the highest estimated diffusivity
and storativity). We obtain an estimated transmissivity 𝐴𝐴 of 10−7 𝑚𝑚2 ∕s, which yields a barometric signal with
a gain
𝐴𝐴 of 3%, which is negligible with respect to our observed signal which gain was around 𝐴𝐴 90%. Thus, the
contribution of horizontal flow from the top layer is indeed negligible compared to that of the bottom one.
Regarding the uncertainty on diffusivity and loading efficiency plotted on Figure 10, it mostly comes from
the uncertainty on phase lag and measured gain (Figure 4). We propagated the uncertainty in the inversion of
the model and plotted it in Figure S4 in Supporting Information S1, showing how uncertainty increases with
increasing diffusivity.
6. Conclusion
In this study, we proposed a new conceptual model for a bi-layer aquifer and applied it to a volcanic aquifer from
Martinique Island. It is able to predict positive phase lag for earth tide while the whole aquifer is confined, which
is unprecedented in the literature. We performed a tidal analysis which enables to distinguish the contributions
of the two identified sources of signal: solid-earth and atmospheric tides. The distinct response to these different
sources yields useful information to understand the aquifer geometry and flow regime. Consequently, we were
able to demonstrate that, starting from a general analytical model, we could use geological observations and
the interpretation of tidal signals to constrain it down to a simpler site-specific model which could be explicitly
inverted. These results were validated by several pumping tests spread across 14 years. We compared our model
to existing ones, showing that keeping it as general as possible at first is an asset when trying to represent particu-
lar geometries.
The diffusion evolution we obtained yields important information on the phenomena that takes place. We
induced a criterion for magnitude-distance selection of earthquakes that impact the aquifer. This criterion
requires that dynamic strains from surface waves are responsible for the observed changes, since they decay
as the square of the distance. This observation brings important information when trying to refine criteria
for distinct phenomena, like water level changes, liquefaction, or, as set forth here, permeability changes.
Secondly, we demonstrated that pumping tests are not innocent in the evolution of aquifer diffusivity, nor
heavy tropical rainfall events, for which we proposed two possible mechanisms to explain the observed tran-
sient decreases in diffusivity, based on effective pressure and water viscosity change linked to colloids and
particles.
Appendix A: Solution to the Diffusion Equation in the Top Layer

To solve
𝐴𝐴 for (13) with boundary conditions
𝐴𝐴 𝐴𝐴 and (15), we define the function
(14)
(A1)
𝑃𝑃 = 𝑝𝑝′ − 𝛾𝛾 ′ 𝑃𝑃𝑎𝑎
We are looking for periodic solutions at a specific frequency

𝐴𝐴 𝐴𝐴𝐴 In complex notation,
𝐴𝐴 𝑃𝑃̃ = 𝑃𝑃 𝑃𝑃−𝑖𝑖𝑖𝑖𝑖𝑖 is the solution of
𝜕𝜕 2 𝑃𝑃̃ 𝑖𝑖𝑖𝑖
(A2) = ′ 𝑃𝑃̃
𝜕𝜕𝜕𝜕2 𝐷𝐷
The boundary conditions are now:
𝜕𝜕 𝑃𝑃̃
(A3) =0
𝜕𝜕𝜕𝜕 |𝑧𝑧=𝑏𝑏′
(A4)
𝑃𝑃̃ (𝑧𝑧 = 0) = 𝜌𝜌𝜌𝜌 ℎ(𝑟𝑟)
̃ − 𝛾𝛾 ′ 𝑃𝑃̃𝑎𝑎
𝐴𝐴 𝑃𝑃̃ is of the form

) (√ ) (√
𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖
(A5)
𝑃𝑃̃ = 𝑃𝑃1 cosh 𝑧𝑧 + 𝑃𝑃2 sinh 𝑧𝑧
𝐷𝐷′ 𝐷𝐷′
(√ )
where √ . (𝐴𝐴 − 4) yields
𝐴𝐴 that 𝐴𝐴1 = 𝜌𝜌𝜌𝜌 ℎ(𝑟𝑟) 𝐴𝐴, and (𝐴𝐴 − 3)
𝐴𝐴 that 𝐴𝐴2 = −𝑃𝑃1 tanh , thus
√ 1+𝑖𝑖 𝑖𝑖𝑖𝑖 ′
𝐴𝐴 =
𝑖𝑖 𝐴𝐴 ̃ − 𝛾𝛾 ′ 𝑃𝑃𝑎𝑎𝑎0 𝑏𝑏
2 𝐷𝐷′
( (√ ) (√ ) (√ ))
𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 ′ 𝑖𝑖𝑖𝑖
(A6)
̃ ̃
( ′
)
𝑃𝑃 (𝑧𝑧) = 𝜌𝜌𝜌𝜌 ℎ(𝑟𝑟) − 𝛾𝛾 𝑃𝑃𝑎𝑎𝑎0 cosh 𝑧𝑧 − tanh 𝑏𝑏 sinh 𝑧𝑧
𝐷𝐷′ 𝐷𝐷′ 𝐷𝐷′
which immediately 𝐴𝐴
yields (16).
Appendix B: Solution to the Diffusion Equation in the Bottom Layer, Adapted From
Wang et al. (2018)
Starting from Equation (18)
( 2 ) ( ) ( )
𝜕𝜕 ℎ̃ 1 𝜕𝜕 ℎ̃ 𝐾𝐾 ′ ̃ 𝛾𝛾 ′ ̃ √ √ 𝛾𝛾 ̃
𝑇𝑇 + − ′ ℎ− ̃
𝑃𝑃𝑎𝑎 (1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄) = 𝑖𝑖𝑖𝑖𝑖𝑖 ℎ − 𝑃𝑃𝑎𝑎
𝜕𝜕𝜕𝜕2 𝑟𝑟 𝜕𝜕𝜕𝜕 𝑏𝑏 𝜌𝜌𝜌𝜌 𝜌𝜌𝜌𝜌
We first solve for the hydraulic head away from the𝐴𝐴well 𝐴∞ (𝑡𝑡), which also follows Equation (18) yet does not
depend
𝐴𝐴 on 𝐴𝐴 :
( ) ( )
𝐾𝐾 ′ 𝛾𝛾 ′ ̃ √ √ 𝛾𝛾 ̃
(B1) − ′ ℎ̃ ∞ − 𝑃𝑃𝑎𝑎 (1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄) = 𝑖𝑖𝑖𝑖𝑖𝑖 ℎ̃ ∞ − 𝑃𝑃𝑎𝑎
𝑏𝑏 𝜌𝜌𝜌𝜌 𝜌𝜌𝜌𝜌
Rearranging the terms, it yields the response:

′
⎛ 𝛾𝛾 ′ + 𝛾𝛾 ⋅ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 1
√ √ ⎞⎟
̃
𝜌𝜌𝜌𝜌 ℎ∞ ⎜ ⎜ ′
𝐾𝐾 (1 + 𝑖𝑖) 𝑄𝑄 tanh((1 + 𝑖𝑖) 𝑄𝑄)
(B2) = ⎟
𝑃𝑃̃𝑎𝑎 ⎜ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖′ 1
1 +
⎟
⎜ √ √ ⎟
⎝ 𝐾𝐾 ′ (1 + 𝑖𝑖) 𝑄𝑄 tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎠
′
Noticing
𝐴𝐴 that 𝐾𝐾𝑖𝑖𝑖𝑖𝑖𝑖
′ (1+𝑖𝑖)
= (1+𝑖𝑖)𝑄𝑄
𝑆𝑆 ′
, it becomes
√
⎛ ′ 𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄 ⎞
⎜ 𝛾𝛾 + 𝛾𝛾 ⋅ ′
𝑆𝑆 tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎟
√ ⎟
𝜌𝜌𝜌𝜌 ℎ̃ ∞ ⎜
(B3) =⎜
𝑃𝑃̃𝑎𝑎
√ ⎟
⎜ 𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄 ⎟
⎜ 1 + 𝑆𝑆 ′ √ ⎟
⎝ tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎠
Now we may solve for the effect of the well on the hydraulic head defined as the opposite of drawdown:
(B4)
Δℎ(𝑟𝑟𝑟 𝑟𝑟) = ℎ(𝑟𝑟𝑟 𝑟𝑟) − ℎ∞ (𝑡𝑡)
Substituting
𝐴𝐴 𝐴𝐴𝑟𝑟) by Δℎ(𝑟𝑟𝑟 𝑟𝑟) + ℎ∞𝐴𝐴(𝑡𝑡) in (10) and inserting
𝐴(𝑟𝑟𝑟 𝐴𝐴 (𝐵𝐵 − 1), we get rid of the source terms and obtain:
( 2 )
𝜕𝜕 Δℎ̃ 1 𝜕𝜕Δℎ̃ 𝐾𝐾 ′ ̃ √ √
(B5) 𝑇𝑇 2
+ − ′ Δℎ(1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄) = 𝑖𝑖𝑖𝑖𝑖𝑖Δℎ̃
𝜕𝜕𝜕𝜕 𝑟𝑟 𝜕𝜕𝜕𝜕 𝑏𝑏
With adapted boundary conditions
(B6)
̃ = ∞) = 0
Δℎ(𝑟𝑟
̃
̃ 𝑤𝑤 , 𝑡𝑡) = ℎ̃ 𝑤𝑤 − ℎ̃ ∞ + 𝑃𝑃𝑎𝑎
(B7)
Δℎ(𝑟𝑟
𝜌𝜌𝜌𝜌
𝜕𝜕Δℎ̃
(B8)
2𝜋𝜋𝜋𝜋𝑤𝑤 𝑇𝑇 = 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖2𝑐𝑐 ℎ̃ 𝑤𝑤
𝜕𝜕𝜕𝜕 |𝑟𝑟=𝑟𝑟𝑤𝑤
From here we can draw a parallel with (Wang et al., 2018, Equations 9–12), which are 𝐴𝐴 like (𝐵𝐵 −
𝐴𝐴 5) to (𝐵𝐵 − 8) once
𝐾𝐾 ′
is substituted to 𝐾𝐾 ′
and ̃ ∞ +𝐴𝐴 𝑃𝑃̃𝑎𝑎 to ℎ̃ ∞.
√ √
𝐴𝐴 𝑏𝑏′
(1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄) 𝐴𝐴 𝐴𝐴
𝑏𝑏′
− ℎ 𝜌𝜌𝜌𝜌
Hence, we define
( ′ )1
(B9) 𝐾𝐾 𝑖𝑖𝑖𝑖𝑖𝑖 2
√ √
𝛽𝛽 = ′
(1 + 𝑖𝑖) 𝑄𝑄 ⋅ tanh((1 + 𝑖𝑖) 𝑄𝑄) +
𝑇𝑇 𝑇𝑇 𝑇𝑇
And by analogy
2
̃ 𝑟𝑟) = − 𝑖𝑖𝑖𝑖𝑖𝑖𝑐𝑐 𝐾𝐾0 (𝛽𝛽𝛽𝛽) ℎ̃ 𝑤𝑤
(B10)
Δℎ(𝑟𝑟𝑟
2𝑇𝑇 𝑇𝑇𝑇𝑇𝑤𝑤 𝐾𝐾1 (𝛽𝛽𝛽𝛽𝑤𝑤 )
Where
𝐴𝐴 𝐴𝐴𝐴𝐴0 and 𝐴𝐴1 are the modified Bessel functions of the second kind, respectively of the zeroth and first order.
Finally,
𝐴𝐴 (𝐵𝐵 − 10) yields𝐴𝐴with (𝐵𝐵 − 𝐴𝐴
7) and (𝐵𝐵 − 3)
√
⎛ ′ 𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄 ⎞
⎜ 𝛾𝛾 − 1 + (𝛾𝛾 − 1) ′
𝑆𝑆 tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎟
√ ⎟
𝑃𝑃̃𝑎𝑎 1 ⎜
(B11) ̃
ℎ𝑤𝑤 = ⋅
𝜌𝜌𝜌𝜌 𝜉𝜉 ⎜⎜
√ ⎟
𝑆𝑆 (1 + 𝑖𝑖) 𝑄𝑄 ⎟
1+ ′
𝑆𝑆
⎜ √ ⎟
⎝ tanh((1 + 𝑖𝑖) 𝑄𝑄) ⎠
With
( )2
𝑟𝑟𝑐𝑐 𝑖𝑖𝑖𝑖𝑖𝑖𝑤𝑤 𝐾𝐾0 (𝛽𝛽𝛽𝛽𝑤𝑤 )
(B12)
𝜉𝜉 = 1 +
𝑟𝑟𝑤𝑤 2𝑇𝑇 𝑇𝑇 𝐾𝐾1 (𝛽𝛽𝛽𝛽𝑤𝑤 )
It is interesting to write these results as a function of adimensional parameters which were studied in the literature
before, namely in (Hsieh et al., 1987; Rojstaczer, 1988). We thus define in addition
𝐴𝐴 to 𝐴𝐴
( )2
𝑟𝑟2𝑤𝑤 𝜔𝜔 𝑟𝑟𝑐𝑐
(B13)
𝑊𝑊 = ; 𝑐𝑐𝑟𝑟 =
2𝑇𝑇 𝑟𝑟𝑤𝑤
And now
𝑊𝑊 𝐾𝐾0 (𝛽𝛽𝛽𝛽𝑤𝑤 )
(B14)
𝜉𝜉 = 1 + 𝑖𝑖 ⋅ 𝑐𝑐𝑟𝑟
𝛽𝛽𝛽𝛽𝑤𝑤 𝐾𝐾1 (𝛽𝛽𝛽𝛽𝑤𝑤 )
( √ )1
(1 + 𝑖𝑖)tanh((1 + 𝑖𝑖) 𝑄𝑄) ′ 2
(B15)
√
𝛽𝛽𝛽𝛽𝑤𝑤 = 𝑊𝑊 √ 𝑆𝑆 + 2𝑖𝑖𝑖𝑖
𝑄𝑄
Appendix C: Analytical Solution of Equations 24 and 25

Starting back from Equation (24)
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤 𝛾𝛾 ′ − 1
=
𝑃𝑃̃𝑎𝑎 𝑆𝑆
√
(1+𝑖𝑖) 𝑄𝑄
1+ ′
√
𝑆𝑆 tanh((1+𝑖𝑖) 𝑄𝑄)
Provided 𝑄𝑄 𝑄 10, which ensures𝐴𝐴 that tanh((1 + 𝑖𝑖) 𝑄𝑄) = 1, an analytical inversion can be performed. If we set
√ √
𝐴𝐴
𝐴𝐴 𝐴𝐴 and 𝐴𝐴 to respectively be the barometric phase lag and gain plotted in Figures 4a–4c, we get
𝐴𝐴
⎛ ⎞
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤
( )
1 − 𝛾𝛾 ′
(C1)
⎜ ⎟
𝜙𝜙 = arg = arg⎜
−𝑃𝑃̃𝑎𝑎 𝑆𝑆 √ ⎟
⎜ 1 + (1 + 𝑖𝑖) 𝑄𝑄 ⎟
⎝ 𝑆𝑆 ′ ⎠
𝑆𝑆
( √ )
(C2)
𝜙𝜙 = arg 1 + ′ (1 − 𝑖𝑖) 𝑄𝑄
𝑆𝑆
Thus,
𝑆𝑆 √
′
𝑄𝑄
tan(𝜙𝜙) = − 𝑆𝑆 √
(C3)
𝑆𝑆
1 + ′ 𝑄𝑄
𝑆𝑆
And finally:
√ 𝑆𝑆 ′ −tan(𝜙𝜙)
(C4)
𝑄𝑄 = ⋅
𝑆𝑆 1 + tan(𝜙𝜙)
𝐴𝐴Once 𝐴𝐴 is known, the gain immediately 𝐴𝐴

yields 𝐴𝐴 ′ ∶
𝑆𝑆 √ ‖
(C5)
‖
𝛾𝛾 ′ = 1 − 𝐺𝐺 ∗ ‖
‖1 + 𝑆𝑆 ′ (1 + 𝑖𝑖) 𝑄𝑄‖
‖
‖ ‖
Following the same process for earth tide, starting from Equation (25):
( √ )
𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤 𝐵𝐵𝐵𝐵𝑢𝑢 ⋅ 𝑆𝑆𝑆𝑆′ (1 + 𝑖𝑖) 𝑄𝑄
=
𝜀𝜀̃
√
1 + 𝑆𝑆𝑆𝑆′ (1 + 𝑖𝑖) 𝑄𝑄
If we
𝐴𝐴 set𝐴𝐴
𝐴𝐴 and 𝐴𝐴 to respectively be the earth tide phase lag and gain plotted in Figures 4b–4d, we get
𝑆𝑆 √
( )
(C6) 𝜙𝜙 = arg 1 + 𝑖𝑖 + 2 ′ 𝑄𝑄
𝑆𝑆
1
tan(𝜙𝜙) =
(C7) 𝑆𝑆 √
1 + 2 ′ 𝑄𝑄
𝑆𝑆
And finally
√ 𝑆𝑆 ′ 1 − tan(𝜙𝜙)
(C8)
𝑄𝑄 = ⋅
𝑆𝑆 2 tan(𝜙𝜙)
‖ 1 − 𝑖𝑖 𝑆𝑆 √ ‖
‖ 2 + 𝑆𝑆 ′ 𝑄𝑄 ‖
‖ ‖
(C9)
𝐵𝐵𝐵𝐵𝑢𝑢 = 𝐺𝐺 ‖ ‖
‖ 𝑆𝑆 √ ‖
‖
′
𝑄𝑄 ‖
𝑆𝑆
‖ ‖
‖ ‖
Data Availability Statement

Piezometric data are available here: https://ades.eaufrance.fr/Fiche/PtEau?Code=1177ZZ0177/PZ2. The bore-
hole national reference is BSS002NPJJ (1177ZZ0177/PZ2). The barometric data have been purchased from
the French meteorological agency (Météo-France) and are subject to a disclaimer of diffusion. Access to these
barometric data is granted to any scientific project organized by a university or research institute as long as the
activity is not for profit and results are submitted for publication. Request to access the data must be addressed
here: https://donneespubliques.meteofrance.fr/?fond=contact&id_produit=93&commande=1. Station number
is 97209004 and further information about it is available here: https://donneespubliques.meteofrance.fr/meta-
donnees_publiques/fiches/fiche_97209004.pdf. Theoretical tidal strain were computed thanks to the SPOTL
program (Agnew, 2012).
Acknowledgments References
This research project, HydroSeisme
Acworth, R. I., Halloran, L. J. S., Rau, G. C., Cuthbert, M. O., & Bernardi, T. L. (2016). An objective frequency domain method for quan-
(2021–2023), was funded by ERC
tifying confined aquifer compressible storage using Earth and atmospheric tides. Geophysical Research Letters, 43(22). https://doi.
Persismo Grant 865411, led by Harsha S.
org/10.1002/2016GL071328
Bhat (ENS), by the initiative Ecosphere
Agnew, D. C. (2012). Spotl: Some programs for ocean-tide loading [SIO Technical Report] (p. 30). Scripps Institution of Oceanography.
Continentale et Cotière (EC2CO) of
Retrieved from http://escholarship.org/uc/item/954322pg
the Institut National des Sciences de
Agnew, D. C. (2015). 3.06—Earth tides. In G. Schubert (Ed.), Treatise on geophysics (2nd ed., pp. 151–178). Elsevier. https://doi.org/10.1016/
l’Univers/Centre National de la Recherche
B978-0-444-53802-4.00058-0
Scientifique (INSU/CNRS), and from
Aki, K., & Richards, P. G. (2002). Quantitative seismology (2nd ed.). University Science Books. Retrieved from www.uscibooks.com
BRGM Sism’Eau Project (An internal
Bailly, C., Fortin, J., Adelinet, M., & Hamon, Y. (2019). Upscaling of elastic properties in carbonates: A modeling approach based on a multiscale
M4 research program). We wish to thank
geophysical data set. Journal of Geophysical Research: Solid Earth, 124(12), 13021–13038. https://doi.org/10.1029/2019JB018391
Emile Gros from BRGM Martinique
Barbosa, N. D., Hunziker, J., Lissa, S., Saenger, E. H., & Lupi, M. (2019). Fracture unclogging: A numerical study of seismically induced viscous
who participated to the pumping test
shear stresses in fluid-saturated fractured rocks. Journal of Geophysical Research, 23(11), 11705–11727. https://doi.org/10.1029/2019jb017984
campaign.
Barker, J. A. (1988). A generalized radial flow model for hydraulic tests in fractured rock. Water Resources Research, 24(10), 1796–1804. https://
doi.org/10.1029/WR024i010p01796
Boeut, S., Fujii, Y., Kodama, J.-I., Fukuda, D., Dassanayake, A., & Alam, B. A. K. M. (2020). Laboratory investigation on the permeability
variation of fractured inada granite by multiple transient axial stress disturbances. Pure and Applied Geophysics, 177(11), 5385–5396. https://
doi.org/10.1007/s00024-020-02565-2
Brajanovski, M., Gurevich, B., & Schoenberg, M. (2005). A model for P-wave attenuation and dispersion in a porous medium permeated by
aligned fractures. Geophysical Journal International, 163(1), 372–384. https://doi.org/10.1111/j.1365-246X.2005.02722.x
Brodsky, E. E., Roeloffs, E., Woodcock, D., Gall, I., & Manga, M. (2003). A mechanism for sustained groundwater pressure changes induced by
distant earthquakes. Journal of Geophysical Research, 108(B8), 10. https://doi.org/10.1029/2002JB002321
Brodsky, E. E., & Prejean, S. G. (2005). New constraints on mechanisms of remotely triggered seismicity at long valley caldera: Triggering
CONSTRAINTS. Journal of Geophysical Research, 110(B4). https://doi.org/10.1029/2004JB003211
Candela, T., Brodsky, E. E., Marone, C., & Elsworth, D. (2014). Laboratory evidence for particle mobilization as a mechanism for permeability
enhancement via dynamic stressing. Earth and Planetary Science Letters, 13, 279–291. https://doi.org/10.1016/j.epsl.2014.02.025
Carlson, F., & Randall, J. (2012). Mlu: A windows application for the analysis of aquifer tests and the design of well fields in layered systems.
Groundwater, 50(4), 504–510. https://doi.org/10.1111/j.1745-6584.2012.00945.x
DeMets, C., Jansma, P. E., Matti, G. S., Dixon, T. H., Farina, F., Bilham, R., et al. (2000). GPS geodetic constraints on Caribbean-North America
plate motion. Geophysical Research Letter, 27(3), 437–440. https://doi.org/10.1029/1999GL005436
Doan, M.-L., Brodsky, E. E., Prioul, R., & Signer, C. (2008). Tidal analysis of borehole pressure: A tutorial (p. 60). University of California.
Duvall, C. R. (1927). Memorandum on the uncertainty in the amplitudes and phase-angles of Fourier terms. Terrestrial Magnitude Atmosphere
Electrical, 32(2), 65–68. https://doi.org/10.1029/TE032i002p00065
Elkhoury, J. E., Brodsky, E. E., & Agnew, D. C. (2006). Seismic waves increase permeability (Vol. 441, p. 4). https://doi.org/10.1038/nature04798
Elkhoury, J. E., Niemeijer, A., Brodsky, E. E., & Marone, C. (2011). Laboratory observations of permeability enhancement by fluid pressure
oscillation of in situ fractured rock. Journal of Geophysical Research, 116(B2), B02311. https://doi.org/10.1029/2010JB007759
Gallagher, A., Fortin, J., & Borgomano, J. (2022). Seismic dispersion and attenuation in fractured fluid-saturated porous rocks: An experimental
study with an analytic and computational comparison. Rock Mechanics and Rock Engineering, 55(7), 4423–4440. https://doi.org/10.1007/
s00603-022-02875-y
Gassman, F. (1951). Uber die Elastizitat Poroser Medien. Vier Der Natur Gesellschaft in Zurich, 1–23.
Hantush, M. S., & Jacob, C. E. (1955). Non-steady radial flow in an infinite leaky aquifer. Eos, Transactions American Geophysical Union, 36(1),
95–100. https://doi.org/10.1029/tr036i001p00095
He, A., Singh, R. P., Sun, Z., Ye, Q., & Zhao, G. (2016). Comparison of regression methods to compute atmospheric pressure and Earth tidal
coefficients in water level associated with Wenchuan Earthquake of 12 May 2008. Pure and Applied Geophysics, 173(7), 18–2294. https://doi.
org/10.1007/s00024-016-1310-3
Hsieh, P. A., Bredehoeft, J. D., & Farr, J. M. (1987). Determination of aquifer transmissivity from Earth tide analysis. Water Resources Research,
23(10), 10–1832. https://doi.org/10.1029/WR023i010p01824
Ingebritsen, S. E., & Manga, M. (2019). Earthquake hydrogeology (Vol. 13).
Kachanov, M., & Abedian, B. (2015). On the isotropic and anisotropic viscosity of suspensions containing particles of diverse shapes and orien-
tations. International Journal of Engineering Science, 94, 71–85. https://doi.org/10.1016/j.ijengsci.2015.05.003
Larochelle, S., Chanard, K., Fleitout, L., Fortin, J., Gualandi, A., Longuevergne, L., et al. (2022). Understanding the geodetic signature of large
aquifer systems: Example of the Ozark plateaus in central United States. Journal of Geophysical Research: Solid Earth, 127(3). https://doi.
org/10.1029/2021JB023097
Leclerc, F. (2014). Déformation active permanente induite par le méga-chevauchement dans l’arc antillais: Apport des complexes récifaux
quaternaires. IPGP.
Maas, C. (1987). Groundwater flow to a well in a layered porous medium: 2. Nonsteady multiple-aquifer flow. Water Resources Research, 23(8),
1683–1688. https://doi.org/10.1029/WR023i008p01683
Manga, M. (2001). Origin of postseismic streamflow changes inferred from baseflow recession and magnitude-distance relations. Geophysical
Research Letters, 28(10), 2133–2136. https://doi.org/10.1029/2000GL012481
Manga, M., Beresnev, I., Brodsky, E. E., Elkhoury, J. E., Elsworth, D., Ingebritsen, S. E., et al. (2012). Changes in permeability caused by
transient stresses: Field observations, experiments, and mechanisms: Dynamic permeability. Reviews of Geophysics, 50(2). https://doi.
org/10.1029/2011RG000382
Manga, M., & Wang, C.-Y. (2015). Earthquake hydrology. In Treatise on geophysics (pp. 305–328). Elsevier. https://doi.org/10.1016/
B978-0-444-53802-4.00082-8
McMillan, T. C., Rau, G. C., Timms, W. A., & Andersen, M. S. (2019). Utilizing the impact of Earth and atmospheric tides on groundwater
systems: A review reveals the future potential. Reviews of Geophysics, 57(2), 281–315. https://doi.org/10.1029/2018RG000630
Méité, D., Rafini, S., Chesnaux, R., & Ferroud, A. (2022). Review of petroleum and hydrogeology equations for characterizing the pressure front
diffusion during pumping tests. Geosciences, 12(5), 201. https://doi.org/10.3390/geosciences12050201
Météo France. (2008). Bulletin climatologique mensuel Martinique—Octobre 2008. Retrieved from https://donneespubliques.meteofrance.fr/
Montgomery, D. R., & Manga, M. (2003). Streamflow and water well responses to earthquakes. Science, 300(5628), 2047–2049. https://doi.
org/10.1126/science.1082980
Muir-Wood, R., & King, G. C. P. (1993). Hydrological signatures of earthquake strain. Journal of Geophysical Research, 98(B12), 22035–22068.
https://doi.org/10.1029/93JB02219
Neuman, S. P. (1972). Theory of flow in unconfined aquifers considering delayed response of the water table. Water Resources Research, 8(4),
1031–1045. https://doi.org/10.1029/WR008i004p01031
Odling, N. E., Perulero Serrano, R., Hussein, M. E. A., Riva, M., & Guadagnini, A. (2015). Detecting the vulnerability of groundwater in
semi-confined aquifers using barometric response functions. Journal of Hydrology, 520, 143–156. https://doi.org/10.1016/j.jhydrol.2014.11.016
Rau, G. C., McMillan, T. C., Andersen, M. S., & Timms, W. A. (2022). In situ estimation of subsurface hydro-geomechanical properties using
the groundwater response to semi-diurnal Earth and atmospheric tides. Hydrology and Earth System Sciences, 26(16), 4301–4321. https://doi.
org/10.5194/hess-26-4301-2022
Roeloffs, E. (1996). Poroelastic techniques in the study of earthquake-related hydrologic phenomena. Advances in Geophysics, 37, 135–195.
https://doi.org/10.1016/S0065-2687(08)60270-8
Roeloffs, E., Sneed, M., Galloway, D. L., Sorey, M. L., Farrar, C. D., Howle, J. F., & Hughes, J. (2003). Water-level changes induced by
local and distant earthquakes at Long Valley caldera, California. Journal of Volcanology and Geothermal Research, 127(3–4), 269–303.
https://doi.org/10.1016/S0377-0273(03)00173-2
Roeloffs, E. A. (1998). Persistent water level changes in a well near Parkfield, California, due to local and distant earthquakes. Journal of
Geophysical Research, 103(B1), 869–889. https://doi.org/10.1029/97JB02335
Rojstaczer, S. (1988). Determination of fluid flow properties from the response of water levels in wells to atmospheric loading. Water Resources
Research, 24(11), 1927–1938. https://doi.org/10.1029/WR024i011p01927
Rojstaczer, S., & Agnew, D. C. (1989). The influence of formation material properties on the response of water levels in wells to Earth tides and
atmospheric loading. Journal of Geophysical Research, 94(B9), 12403. https://doi.org/10.1029/JB094iB09p12403
Rojstaczer, S., & Riley, F. S. (1990). Response of the water level in a well to Earth tides and atmospheric loading under unconfined conditions.
Water Resources Research, 26(8), 1803–1817. https://doi.org/10.1029/WR026i008p01803
Rojstaczer, S., Wolft, S., & Micheli, R. (1995). Permeability enhancement in the shallow crust as a cause of earthquake-induced hydrological
changes. Nature, 373(6511), 3239. https://doi.org/10.1038/373237a0
Shi, Y., Liao, X., Zhang, D., & Liu, C. (2019). Seismic waves could decrease the permeability of the shallow crust. Geophysical Research Letters,
46(12), 6371–6377. https://doi.org/10.1029/2019GL081974
Sun, X., Shi, Z., & Xiang, Y. (2020). Frequency dependence of in situ transmissivity estimation of well-aquifer systems from periodic loadings.
Water Resources Research, 56(11). https://doi.org/10.1029/2020WR027536
Theis, C. V. (1935). The relation between the lowering of the Piezometric surface and the rate and duration of discharge of a well using
ground-water storage. Transactions - American Geophysical Union, 16, 519–524. https://doi.org/10.1029/tr016i002p00519
Valois, R., Rau, G. C., Vouillamoz, J.-M., & Derode, B. (2022). Estimating hydraulic properties of the shallow subsurface using the
groundwater response to Earth and atmospheric tides: A comparison with pumping tests. Water Resources Research, 36(5). https://doi.
org/10.1029/2021WR031666
Vittecoq, B., & Brugeron, A. (2008). Recherche d’eau souterraine à Schoelcher – Fond Lahaye (RP-55728-FR. BRGM. 242. Retrieved from
http://ficheinfoterre.brgm.fr/document/RP-55728-FR
Vittecoq, B., Fortin, J., Maury, J., & Violette, S. (2020). Earthquakes and extreme rainfall induce long term permeability enhancement of volcanic
island hydrogeological systems. Scientific Reports, 10(1), 20231. https://doi.org/10.1038/s41598-020-76954-x
Vittecoq, B., Reninger, P.-A., Lacquement, F., Martelet, G., & Violette, S. (2019). Hydrogeological conceptual model of andesitic water-
sheds revealed by high-resolution heliborne geophysics. Hydrology and Earth System Sciences, 23(5), 2321–2338. https://doi.org/10.5194/
hess-23-2321-2019
Vittecoq, B., Reninger, P. A., Violette, S., Martelet, G., Dewandel, B., & Audru, J. C. (2015). Heterogeneity of hydrodynamic properties and
groundwater circulation of a coastal andesitic volcanic aquifer controlled by tectonic induced faults and rock fracturing – Martinique island
(Lesser Antilles – FWI). Journal of Hydrology, 529, 1041–1059. https://doi.org/10.1016/j.jhydrol.2015.09.022
Wang, C., Liao, X., Wang, L., Wang, C., & Manga, M. (2016). Large earthquakes create vertical permeability by breaching aquitards. Water
Resources Research, 52(8), 5923–5937. https://doi.org/10.1002/2016WR018893
Wang, C.-Y., Doan, M.-L., Xue, L., & Barbour, A. J. (2018). Tidal response of groundwater in a leaky aquifer—Application to Oklahoma. Water
Resources Research, 15(10), 8019–8033. https://doi.org/10.1029/2018wr022793
Wang, C.-Y., & Manga, M. (2010). Hydrologic responses to earthquakes and a general metric. Geofluids. https://doi.org/10.1111/j.1468-8123.
2009.00270.x
Wang, C.-Y., & Manga, M. (2015). New streams and springs after the 2014 Mw6.0 south Napa earthquake. Nature Communications, 6(1), 7597.
https://doi.org/10.1038/ncomms8597
Wang, C.-Y., Wang, C.-H., & Kuo, C.-H. (2004). Temporal change in groundwater level following the 1999 (Mw = 7.5) Chi-Chi earthquake,
Taiwan. Geofluids, 4(3), 210–220. https://doi.org/10.1111/j.1468-8123.2004.00082.x
Wang, H., Xin, J., Zheng, X., Li, M., Fang, Y., & Zheng, T. (2020). Clogging evolution in porous media under the coexistence of suspended
particles and bacteria: Insights into the mechanisms and implications for groundwater recharge. Journal of Hydrology, 582, 124554. https://
doi.org/10.1016/j.jhydrol.2020.124554
Wang, H. F. (2000). Theory of linear poroelasticity with applications to geomechanics and hydrogeology. Princeton University Press.
Warren, J. E., & Root, P. J. (1963). The behavior of naturally fractured reservoirs. Society of Petroleum Engineers Journal, 3(03), 245–255. https://
doi.org/10.2118/426-PA
Westercamp, D., Pelletier, B., Thibault, P. M., & Traineau, H. (1990). Carte géologique de la Martinique au 1/50000ème. Editions BRGM.
Retrieved from http://infoterre.brgm.fr/formulaire/telechargement-cartes-geologiques-drom
Xiang, Y., Sun, X., & Gao, X. (2019). Different coseismic groundwater level changes in two adjacent wells in a fault-intersected aquifer system.
Journal of Hydrology, 578, 124123. https://doi.org/10.1016/j.jhydrol.2019.124123
Xue, L., Li, H.-B., Brodsky, E. E., Xu, Z.-Q., Kano, Y., Wang, H., et al. (2013). Continuous permeability measurements record healing inside the
Wenchuan Earthquake fault zone, 340, 6. https://doi.org/10.1126/science.1237237
Yan, R., Woith, H., & Wang, R. (2014). Groundwater level changes induced by the 2011 Tohoku earthquake in China mainland (Vol. 16).
Zhang, H., Shi, Z., Wang, G., Sun, X., Yan, R., & Liu, C. (2019). Large earthquake reshapes the groundwater flow system: Insight from the water-level
response to Earth tides and atmospheric pressure in a deep well. Water Resources Research, 13. https://doi.org/10.1029/2018WR024608
Zhang, S., Shi, Z., & Wang, G. (2019). Comparison of aquifer parameters inferred from water level changes induced by slug test, Earth tide and
earthquake – A case study in the three Gorges area. Journal of Hydrology, 10, 124169. https://doi.org/10.1016/j.jhydrol.2019.124169

Earthquakes_and_heavy_rainfall_influence_on_aquife

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Earthquakes_and_heavy_rainfall_influence_on_aquife

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Earthquakes_and_heavy_rainfall_influence_on_aquife

Uploaded by

Copyright:

Available Formats

RESEARCH ARTICLE Earthquakes and Heavy Rainfall Influence on Aquifer

2. The Martinique Fond Lahaye Aquifer

2.2. Data Presentation

2.3. Pumping Tests

𝐴𝐴with ℎ̃ 𝑤𝑤𝑤𝑤𝑤 the complex amplitude. We recover each complex amplitude

𝜌𝜌𝜌𝜌 ℎ̃ 𝑤𝑤𝑤1 = 𝑍𝑍𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏12 ⋅ 𝑃𝑃̃𝑎𝑎𝑎1 + 𝑍𝑍𝜀𝜀𝜀12 ⋅ 𝜀𝜀̃1

This linear system is solved as:

𝑃̃𝑎,1 ⋅ ℎ̃ 𝑤,2 − 𝑃̃𝑎,2 ⋅ ℎ̃ 𝑤,1

3.2. Tidal Analysis Results

𝐴𝐴 to 𝐴𝐴 = |𝑍𝑍𝜀𝜀𝜀𝜀𝜀2∕𝑆𝑆2 | [𝑃𝑃 𝑃𝑃∕𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛], because

4. Tidal Response of a Confined Bi-Layer Aquifer

where 𝐴 is the hydraulic head in the bottom𝐴𝐴layer, 𝐴𝐴 the hydrogeological stor-

loading efficiency of the bottom𝐴𝐴layer (−)

where 𝐴𝐴𝑐𝑐 is the well casing inner radius

 ℎ(𝑟𝑟𝑤𝑤 , 𝑡𝑡) = ℎ𝑤𝑤 (𝑡𝑡)

4.2. Top Layer Response: Computation of the Vertical Leakage

4.3. Whole Aquifer Response: Computation of the Head in the𝑨𝑨Well 𝒉𝒉

are controlled by the same four adimensional parameters:

4.4. Physical Meaning of the Four Adimensional Parameters

 n intermediate frequency regime

4.5. Model Constraint

𝐴𝐴 of 𝐴𝐴𝐴 𝐴𝐴 ′ 𝐾𝐾𝑢𝑢′ ∕𝐵𝐵𝐵𝐵𝑢𝑢 and S/S’ with the hypothesis

From Gassmann's equation (Gassman, 1951), we know that

which is consistent with our case where

4.6. Model Inversion

presented on Figure 10.

Figure 10. Evolution of (a) top layer vertical Diffusivity

4.7. Validation With Pumping Tests

Figure 11. Evolution of (a) top layer vertical Diffusivity

𝐴𝐴 is 2.4 ⋅ 10−4 𝑚𝑚∕s (Table 2)

5.1.2. Extreme Rainfall Events

A second phenomenon has been identified in a previous study in Marti-

where 𝐴𝐴 is the dynamic viscosity of water with a volume fraction

5.1.3. Effect of Pumping in the Aquifer

5.3. Impact of Oceanic Tide

5.4. Comparison With Existing Models

5.5. Validity of Our Hypotheses and Uncertainty Assessment

Appendix A: Solution to the Diffusion Equation in the Top Layer

We are looking for periodic solutions at a specific frequency

The boundary conditions are now:

𝐴𝐴 𝑃𝑃̃ is of the form

Rearranging the terms, it yields the response:

With adapted boundary conditions

Appendix C: Analytical Solution of Equations 24 and 25

𝐴𝐴Once 𝐴𝐴 is known, the gain immediately 𝐴𝐴

Data Availability Statement

You might also like

ℎ(𝑟𝑟𝑤𝑤 , 𝑡𝑡) = ℎ𝑤𝑤 (𝑡𝑡)

n intermediate frequency regime