Environmental Earth Sciences (2022) 81:381

https://doi.org/10.1007/s12665-022-10492-9

ORIGINAL ARTICLE

Assessment of the regional subsidence in the lacustrine zone

of Mexico City using a geostatistical model
M. C. Madrigal1 · E. Botero1 · C. Díaz‑Ávalos2

Received: 5 October 2021 / Accepted: 15 June 2022 / Published online: 18 July 2022
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022

Abstract
In the last 100 years, the overexploitation of the groundwater resources underlying the lacustrine deposits of Mexico City
has trigged regional subsidence that manifests itself through the sinking of the ground surface, damaging infrastructure and
public services. In this work, geostatistical tools are used to analyze and assess the evolution of this phenomenon in the city,
by considering space–time data from surface benchmarks located in the lacustrine zone. The database is composed of 206
spatial features and 12-time points, for a total of 24 years of monitoring between 1983 and 2007. Here a full grid space–time
layout (STF) of the R spacetime package was used, and since this phenomenon is not a stationary process as it presents a
trend over time, the spatiotemporal variogram was determined from the stochastic residual function of the process. Marginal
and pooled variograms were also determined as initial values to fit the variograms models using the R gstat package. Results
show that the separable variogram model was the one that best represented the spatial and temporal correlation of the phe-
nomenon in the area of study. Using this geostatistical model, ground elevations and subsidence rates were predicted in the
benchmarks locations for the period 2010–2030. The validation process consisted of comparing direct measurements made
in 2016 and the ones obtained from the model for the same year. For the year 2030, the maximum cumulative subsidence
value was 13 m, and the highest mean rate was 25.06 cm/year near the Mexico City International Airport.

Keywords Geostatistics · Space–time models · Kriging · Regional subsidence · R

Introduction
The subsoil of the Valley of Mexico is divided into three
geotechnical zones which are the Hill zone (Zone I), the
Transition zone (Zone II), and the Lake zone (Zone III). The
Hill zone is mainly composed of a surface layer of lava flows
or volcanic tuffs (Shapiro et al. 2001; Arroyo et al. 2013).
On the other hand, the Transition zone presents alluvial
sandy and silty layers interbedded with occasional clay lay-
ers (Shapiro et al. 2001; Arroyo et al. 2013). However, most
of Mexico City is settled in the Lake zone in which large
amounts of volcanic ash and other pyroclastic materials were
deposited (Marsal 1969; Mesri et al. 1975; Romo et al. 1988;
* E. Botero
EBoteroJ@iingen.unam.mx
1
Instituto de Ingeniería, Universidad Nacional Autónoma de
México (UNAM), Mexico City, Mexico
2
Instituto de Investigaciones en Matemáticas Aplicadas y
en Sistemas, Universidad Nacional Autónoma de México
(UNAM), Mexico City, Mexico
Santoyo 2005; Ovando et al. 2007; Ossa et al. 2019). Over
time, chemical degradation of these materials formed clays
and clayey silts that are characterized by their high-water
content, compressibility, and low shear strength (Bard et al.
1988; Ovando et al. 2003, 2007, 2013; Ossa et al. 2019).
The soil profile of the lacustrine area within Mexico City is
composed of a dried clay surface crust and alluvial deposits,
followed by a stratum of very soft clay called the Upper Clay
Formation (UCF) with insertions of sandy silt lenses, silty
sands, volcanic glass, and fossils. Underneath this stratum,
there is a stratum made up of extremely compact sandy silts
at a depth of approximately 40 m, which rests on top of a soft
layer of clay called the Lower Clay Formation (LCF) which,
on average, is 10 m thick (Romo et al. 1988; Ovando et al.
2007, 2013; Ossa et al. 2019). At a depth of approximately
50 m, an upper granular aquifer is made up of Quaternary
alluvial deposits, volcaniclastic and pyroclastic rocks, and
Plio-Quaternary basaltic-andesitic rocks are found, which
for practical purposes, provide drinking water for most of
the population (Hernández-Espriu et al. 2014). Ground-
water depth ranges between 60 and 170 m, while saturated

13
Vol.:(0123456789)

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

381 Page 2 of 16
thickness might exceed 800 m in some areas (Hernández-
Espriu et al. 2014). Extraction wells are screened within
the upper-most 300–400 m of this unit (Herrera et al. 1989,
Hernández-Espriu et al. 2014).
The exploitation of the aquifer underlying the clay depos-
its and formed by two units, one granular and one of the frac-
tured volcanic rocks, began in the mid-nineteenth century
to supply water to the city's population (González-Morán
et al. 1999, Ovando et al. 2003; Santoyo et al. 2005; Ovando
et al. 2007). In 1940, abstraction was 4 ­m3/s and by 1990, it
reached 35 ­m3/s (González-Morán et al. 1999). Currently,
water consumption in the basin amounts to 61.6 ­m3/s, out
of which 66% are supplied by wells. Most of the extractions
take place from a regional complex of Quaternary–Ter-
tiary alluvial, pyroclastic, and fractured aquifers (Hernán-
dez-Espriu et al. 2014). However, the intense exploitation
of the groundwater resources from which an average of
203 × ­106 ­m3 of water is extracted each year, of which only
156 × ­106 ­m3 is naturally recharged (Arroyo et al. 2013),
has triggered a regional consolidation of lacustrine deposits
that is manifested by the lowering of the land surface as
voids in the soil are reduced and forces between soil parti-
cles increase. The first author to establish the relationship
between population growth, the amount of water extracted
from the aquifer, and subsidence rates in Mexico City was
reported by Carrillo (1948), and later, Marsal and Mazari
(1959) and Ovando et al. (2013).
From 1900 to 1920 the rate of subsidence in downtown
Mexico City was 3 cm/year, then it increased to 13 cm/
year in the 1940s and reached 26 cm/year in the early
1950s (Tamez et al. 1995, 1997; Ovando et al. 2007). Then,
between 1956 and 1967 the rate of subsidence decreased
to 5 cm/year because extraction of water from the wells in
downtown Mexico City was banned (Ovando et al. 2003,
2007; Aguilera 2013). However, in the late 1970s and early
1980s, new wells were put in operation in the surround-
ing area of the city and the sinking rate increased again to
10 cm/year in downtown Mexico City and at some sites near
the new extraction wells, it even exceeds 30 cm/year (Mazari
1996; Ovando et al. 2003, 2007; Aguilera 2013). In the last
100 years, this phenomenon has led to having cumulative
ground subsidence of more than 8 m in some areas of the
lacustrine zone of the Valley of Mexico compared to a refer-
ence point outside the lake zone (Ovando et al. 2003, 2013;
Arroyo et al. 2013).
The importance of analyzing the behavior of regional
subsidence in the lacustrine zone of Mexico City lies in
the fact that it can affect the functionality and safety of the
city's public services and infrastructure when a significant
accumulation of settlements occurs. For instance, the eastern
part of the city has experienced the reactivation of faults
and fractures because groundwater depletion and differential
land subsidence generate horizontal tensile stresses that can
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Environmental Earth Sciences (2022) 81:381
reach up to several meters, causing damage to gas stations,
underground storage tanks, and oil pipelines (Hernández-
Espriu et al. 2014). On the other hand, the drainage system
is strongly affected by subsidence because it modifies the
gradient, and in some cases, it can cause the inversion of
its slope (Lesser and Cortés 1998). In addition, this phe-
nomenon modifies the static and dynamic properties of the
ground, causing a change in the response of the soil deposit
during a seismic event. Indirect consequences also include
a decrease in water resources and an increase in flood risk
and water contamination, all endangering human lives and
resulting in heavy financial burdens for local government
administrations (Cabral-Cano et al. 2008; Ortiz-Zamora and
Ortega-Guerrero 2010; Chaussard et al. 2014). Because of
this and given that subsidence in Mexico City is one of the
most representative cases of this type in the world, different
studies have been carried out to assess the evolution of this
phenomenon and mitigate its effects.
In 1976, the General Directorate of Construction and
Hydraulic Operation (DGCOH, in Spanish) of the Fed-
eral District government began the installation of surface
benchmarks (topographic references whose elevation and
location coordinates are known) to understand the behavior
of regional subsidence over time, provide solutions to the
problems of hydraulic systems in the Valley of Mexico and
assist in the design of future projects (SIMOH 2015). As a
result of this work, the evolution of the benchmarks in the
period 1984–2002 and a map with the cumulative subsid-
ence of the entire area of Mexico City are presented in the
Integrated Water Resources Management Program (SAC-
MEX 2005). Lesser and Cortés (1998) prepared a historical
analysis and simulated subsidence along the 17 main drain-
ages of Mexico City, including the Grand Canal and the
Churubusco River, among others, using a linear statistical
method, a logarithmic regression, and the application of a
mathematical model developed by Cruickshank et al. (1979).
In this study, the annual mean rate of subsidence was deter-
mined, obtaining maximum values of 30 cm/year within the
limits of Mexico City and Netzahualcoyotl City; from 20 to
25 cm/year in the Mexico City International Airport; around
10 cm/year in downtown Mexico City; between 2 and 5 cm/
year in the Azcapotzalco area, and approximately 15 cm/year
in Xochimilco and Canal Chalco area.
Recent researchers such as Cabral-Cano et al. (2008,
2011) and Osmanoǧlu et al. (2011) focused on spatio-tem-
poral variations of subsidence rates based on space-geodetic
techniques such as Radar Interferometry (InSAR), Persistent
Scatter Interferometry (PSI) and global positioning system
(GPS) analysis (Hernández-Espriu et al. 2014).
While some studies focused on measuring the ground
subsidence for several sites in Mexico City, others were
developed to analyze the effects of this phenomenon on the
static and dynamic properties of lacustrine deposits and their
Environmental Earth Sciences (2022) 81:381 Page 3 of 16 381

seismic response during a seismic event. Jaime (1988) iden-

tified that the water content, strength, compressibility, and
dynamic properties of the soil underlying Mexico City are
time-dependent because of water extraction (Arroyo et al.
2013). The works of Ovando et al. (2003, 2007, 2013) were
based on the model of Yin and Graham (1994, 1996) to pro-
pose expressions to determine the deformations generated
during the one-dimensional consolidation process in the soil
mass. On the other hand, Avilés and Pérez-Rocha (2010)
proposed an empirical model that extrapolates historical
data of the elevations of the metropolitan area of Mexico
City obtained from benchmarks, to evaluate and predict the
evolution of the natural period of soil vibration T ­ s and the
depth of Deep Deposits between the years 2005 and 2055.
In addition, Arroyo et al. (2013) presented field evidence
about the effect of the subsidence phenomenon on the natu-
ral period of soil vibration ­Ts and developed an empirical
model to predict the value of this parameter at different sites
in Mexico City for the years 2010 and 2050.
Hence, it can be evidenced that there is not a specific
theoretical or empirical method to analyze the subsidence
phenomenon in the lacustrine area of Mexico City, since it
is a complex process in which there is an interaction between
hydraulic, geotechnical, geological, and hydrogeological fac-
tors. The purpose of this work is to assess the behavior in
space and time of the surface benchmarks located in the
Lake zone of Mexico City and to analyze the evolution of
the regional subsidence phenomenon from a statistical point
of view using a space–time model.
Space–time data analysis methods
Geostatistics is a set of statistical techniques used for the
analysis of stochastic processes whose domain D is continu-
ous in a subset of ℝd . If we denote the random variable of
interest by Z and s as an arbitrary point in D then the sto-
chastic process of interest is denoted as {Z(s) ∶ s𝜖D}, and
represents the set of possible observations of Z(s), which
form a random field (Cressie 1993). One of the usual tasks in
geostatistical data analysis is the prediction of Z at unvisited
locations s0 . In general, the geostatistical analysis consists
of the following four stages: exploratory analysis, structural
analysis, spatial modeling, and prediction using krige.
The set of methodologies used to analyze space–time data
is based on well-established geostatistical techniques, where
time is considered as an additional spatial dimension. Care
must be taken for space–time predictions as the time dimen-
sion behaves differently than the dimensions in ℝd.
In the spacetime package of the R statistical program (R
Core Team 1993; RStudio Team 2015), four grid formats
for spatiotemporal data are included, but in this work, the
full grid space–time layout (STF) is used. This grid repre-
sents a combination of spatial si , i = 1, … , n and temporal
Fig. 1  Full grid layout ( modified from Pebesma 2012)
tj , j = 1, … , m attributes for a total of zk stored observations,
with k = 1, … ., nm , which represent all possible locations
in the spatiotemporal grid of the phenomenon of interest
(Fig. 1).
Spatio‑temporal models
Consider a spatio-temporal random field Z(s, t) defined in a
spatial domain D{and ( a temporal
) ( domain
)} T, for which there
is a sample z = {(z s1 , t)1 , … (, z sn ,)tn of observations } at
known locations s1 , t1 , … , sn , tn 𝜖S × T ⊆ D × [0, ∞) .
(Gräler et al. 2016). To perform estimations and predic-
tions of the variable Z at unobserved locations, the ran-
dom field Z is commonly assumed to be stationary and
spatially isotropic, so it can be characterized through the
mean μ and a covariance function Cst . Under this assump-
tion, the space–time covariance depends only on the sepa-
rating distances across space h ∈ ℝ2 ≥ 0 and time u ∈ ℝ+ ,
this is, C(h, u) = Cov(Z(s, t), Z(̃s, ̃t)), where h = ‖s − s̃‖ and
u = ‖t − ̃t‖ are the spatial and temporal distances of a pair
of points (s, t),(̃s, ̃t)𝜖S × T . The spatio-temporal variograms
𝛾st (h, u) = Cst (0, 0) − Cst (s, t) are often used to characterize a
phenomenon Z, since unlike the covariance, is not necessary
to estimate the mean of Z to compute 𝛾st (h, u) (Gräler et al.,
2016). The variogram models implemented in the R gstat
package make it possible to relate objects to spatiotemporal
characteristics and to estimate these in space and time.
A stochastic process Z(s, t) has a fully separable covari-
ance function if for every pair of points (s, u) ∈ D;(t, v) ∈ ℝ+
Cov[Z(s, t), Z(u, v)] ≡ Cov(s, u)Cov(t, v) (1)
If the space and the time components are stationary,
Cov(s, u)Cov(t, v) = CS (||s − u||)CT (|t − v|) (2)

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

381 Page 4 of 16 Environmental Earth Sciences (2022) 81:381

Under the stationarity assumption, the separable vari- sillst = k∙sills ∙ sillt + sills + sillt . (6)
ogram is obtained as
Bilonick (1988) proposed the sum-metric model (Eq. 7),
1
𝛾(s, u; t, v) = C(s, s; t, t) + C(u, u: v, v) − C(s, u;t, v) (3) an extension of the separable variogram model, using zonal
2
and geometric anisotropy to resolve the differences between
There are several separable variogram model. The sepa- spatial and temporal variability (Snepvangers et al. 2003).
rable variogram model of Eq. 4 has a great computational (√ )
advantage when each spatial object has an observation at 𝛾sm (h, u) = 𝛾s (h) + 𝛾t (u) + 𝛾joint h2 + (k ∙ u)2 (7)
each temporal instance, i.e., when the data are organized
in a full grid layout (Pebesma 2012). On the other hand,
where, 𝛾s (h), 𝛾t (u) and 𝛾joint are the spatial, temporal, and
it has the simplest form for spatio-temporal models, where
joint variograms, respectively, and k is the anisotropy.
the dependence on the spatial coordinates is separated from
the dependence on the temporal coordinates (De Iaco et al.
2011).
Spatio‑temporal analysis of the regional
(
𝛾sep (h, u) = sill ∙ 𝛾 s (h) + 𝛾 t (u) − 𝛾 s (h)𝛾 t (u)
)
(4) subsidence in Mexico City

This model extends to the product-sum variogram model
(Eq. 5) developed by De Cesare et al. (2001) and De Iaco
et al. (2001) where, unlike the previous one, they considered
non-separable variograms to represent the spatiotemporal
variability of the data.
( ) ( )
𝛾ps (h, u) = k ∙ sillt + 1 𝛾s (h) + k ∙ sills + 1 𝛾t (u) − k𝛾s (h)𝛾t (u)
(5)
Here, 𝛾s (h) and 𝛾t (u) are the spatial and temporal vario-
grams respectively, and k is a positive parameter that defines
the joint sill (Eq. 6),sillst , in terms of the spatial sills and
temporal sillt.
Description of the database
The General Directorate of Hydraulic Construction and
Operation (DGCOH) of the Federal District government,
today Mexico City Water Bureau (SACMEX, in Spanish),
began in 1976 with the installation of surface benchmarks
to understand the evolution of subsidence and to provide
solutions to the problems of hydraulic systems in the Valley
of Mexico (SIMOH 2015).
The monitoring system consists of 1931 surface bench-
marks distributed in the Hill zone (Zone I), the Transition
zone (Zone II), and the Lake zone (Zone III), leveled during

Fig. 2  Distribution of superficial benchmarks in the Valley of Mexico installed by the General Directorate of Hydraulic Construction and Opera-
tion (DGCOH) of the Federal District government, today Mexico City Water Bureau (SACMEX, in Spanish)

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2022) 81:381
1983–2016 with a reading every two years, approximately
(Fig. 2). However, it is important to mention that these data
have missing ground elevation records. The benchmarks
have as a reference point the Atzacoalco benchmark, which
is located at an elevation of 2245.008 m above sea level
(masl) fixed on the rocky body of the Sierra de Guadalupe,
outside the lacustrine zone (SIMOH 2015) (Fig. 2).
To perform the statistical analysis and develop a spa-
tial–temporal model, the information provided was reviewed
and the database was built considering the measurements of
the following superficial references: (1) benchmarks located
in the Lake zone because it is where the regional subsidence
phenomenon manifests itself, and (2) also those that have
1983 as the initial year of monitoring and presented at most
two missing records in non-consecutive years. These missing
readings were obtained from a linear interpolation consider-
ing measurements trend. This process was performed to have
a full space–time grid layout (STF) of observations, where
each spatial point si , i = 1, … , n, has a measurement in each
one of the monitoring years tj , j = 1, … , m (Fig. 1). Finally,
Fig. 3  Ground elevation measurements (masl) between 1983 and 2007 where the analyzed benchmarks are located
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 5 of 16 381
once these benchmarks were selected, those that showed an
erratic behavior during the monitoring time, i.e., that instead
of showing a decrease in the elevation records due to the
sinking ground surface, they showed an increase, were elimi-
nated. Some factors related to this unreliable behavior are
associated with uncalibrated equipment, errors in the meas-
urements, long periods between leveling surveys, and uncer-
tainty about the stratum where the benchmark was installed.
At the end of this data review, 206 spatial features
remained (Fig. 2), each with thirteen-time points (1983,
1985, 1987, 1989, 1992, 1994, 1996, 1998, 2000, 2002,
2005, 2007, 2016) for a total of 33 years of monitoring.
The minimum distance between them is 112.25 m and the
maximum distance is 2,131.32 m. It should be noted that
the time points of the year 2016 were not considered in the
space–time analysis, since they were used to validate the
ground elevations determined with the geostatistical model.
Figure 3 shows ground elevation measurements in masl
for each monitored year at the points where the 206 bench-
marks are located in Zone III. In this figure the readings are
13
381 Page 6 of 16
Fig. 4  Subsidence rate of each
superficial benchmark during
1983–2007
presented in a scale of colors to show, spatially, the sinking
of the ground as the years go by. Green points correspond
to the highest elevations, and the red ones, to the lowest
ones. In addition, the mean rate of subsidence in centimeters
per year (cm/year) was determined for each benchmark and
the Jenks Natural Breaks classification method (Jenks and
Caspall 1971) was used to organize them into 6 classes:(1)
1.1–5.9 cm/year, (2) 5.9–11.1 cm/year, (3) 11.1–16.7 cm/
year, (4) 16.7–23.1 cm/year, (5) 23.1–28.8 cm/year and (4)
28.8–35.1 cm/year (Fig. 4). This method reduces the vari-
ance within classes and maximizes it between classes.
Figure 4 shows that the highest mean rates (red points) are
recorded in the benchmarks located near Mexico City Inter-
national Airport. On the other hand, the lowest rates (black
points) are recorded in the surrounding area of the lacustrine
zone. Between 1983 and 2007, the lowest and highest mean
rates were 1.1 and 35.1 cm/year, respectively. In the bench-
mark P(S04E07)01 where the highest subsidence rate was
recorded, 8.41 m cumulative subsidence occurred. This ref-
erence point is located in the Iztapalapa locality of Mexico
City. Finally, subsidence rate contour lines were obtained by
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Environmental Earth Sciences (2022) 81:381
performing a kriging interpolation using a 100 × 100 m grid
(Fig. 5). In this figure red zones correspond to areas where
subsidence rates are higher.
Currently, besides the conventional leveling of surface
benchmarks, the use of satellite images has also allowed
analyzing the sinking of Mexico City providing a better
spatial resolution and a general overview of the phenom-
enon. Researchers such as Cabral-Cano et al. (2008, 2011)
Osmanoǧlu et al. (2011) and Hernández-Espriu et al. (2014)
have used this data to study the variations in subsidence
rates and land deformation. Overall, the subsidence rate
spatial pattern and the location of the maximum rate values
in the eastern of Mexico City closely to the boundaries of
the ex-Texcoco Lake, presented in the subsidence rate maps
of these studies, agree with the one shown in Fig. 5 for the
1983–2007 period.
Figure 6 gives information about the evolution of ground
elevation measurements during 24 years of monitoring
within the subsidence rate classes (Fig. 4). In this figure,
each line represents the readings recorded in each topo-
graphic reference in the twelve-time points (1983, 1985,
Environmental Earth Sciences (2022) 81:381 Page 7 of 16 381

Fig. 5  Mean subsidence rate

map for the period 1983–2007

1987, 1989, 1992, 1994, 1996, 1998, 2000, 2002, 2005 The spatiotemporal process Z is composed of two func-
2007). It is important to underline that only a ground eleva- tions (Eq. 8), a deterministic one that represents the mean
tion measurement was recorded in each considered year. In behavior of the phenomenon ( m ) and a stochastic residual
addition, this figure gives information about the behavior of (𝜖 ) one.
subsidence rates in each class and the difference between ( ) ( )
them. On the other hand, this figure shows that the regional Z si , ti = m si , ti + 𝜖(si , ti ) (8)
subsidence phenomenon is a non-stationary process as it
Since the regional subsidence phenomenon is not a sta-
presents a trend over time.
tionary process as it presents
( a trend
) over time, the spati-
otemporal variogram 𝛾 si , sj , ti , tj was determined from the
Structural analysis
stochastic residual function 𝜖 of the process as follows:
To determine the behavior of an attribute of interest ( ) [
1 ( ( ) ( ))2 ]
z = {z(s, t)|s𝜖S, t𝜖T} defined in a geographic domain S( ⊂ ℝ2) 𝛾 si , sj , ti , tj = E 𝜖 si , ti − 𝜖 sj , tj (9)
2
and a time interval T ⊂ ℝ, at a spatiotemporal point s0 , t0
where z was not measured, Under the assumptions of stationarity, an estimate of the
( ) the observations of the phenom-
enon Z at points si i are taken into account. To predict
, t spatiotemporal variogram can( be obtained
) by calculating the
( )
z s0 , t0 it is assumed that z is a spatiotemporal realization empirical semi-variogram ̂𝛾 hS , hT from the observations
of the process Z, so it is required to characterize, both spa- as it is shown in Eq. 10.
tially and(temporally,
) the variability of the phenomenon to
predict Z s0 , t0 .

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

381 Page 8 of 16 Environmental Earth Sciences (2022) 81:381

Fig. 6  Evolution of ground elevations between 1983 and 2007 in the lacustrine zone

N (hS ,hT ) [ was assumed that any two temporal consecutive observations
( ) 1 ∑ ( ( ))2 ]
̂
𝛾 hS , hT = ( ) 𝜖(s, t) − 𝜖 s + hS , t + hT have the same temporal distance.
2N hS , hT i=1 In addition, to observe the differences between the semi-
(10) variances in the spatial and temporal domains S and T,
where, hS and hT are( the distances
) in space S and time marginal semi-variograms were also obtained (Fig. 8). In
T, respectively, and N hS , hT correspond to the number of general, low semivariance values are observed at a short dis-
pairs. tance because points that are closer together are more alike
A 2D plot and a 3D wire plot of the empirical spatiotem- and as pairs of locations become farther apart, they are more
poral semi-variogram is shown in Fig. 7. This semi-vario- dissimilar and semivariance values increase. A comparison
gram was determined based on ground-level readings from of the marginal semi-variograms shows that the spatial vari-
each considered benchmark. In this figure, the semivariance ability is larger than the temporal one. Furthermore, it can be
values vary between 0 (purple color) and 9 (yellow color), observed that the spatial semi-variogram reaches an eventual
approximately. It is important to notice that in this case, it sill whereas the temporal one, does not. This is an indication

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2022) 81:381 Page 9 of 16 381

Fig. 7  Empirical spatiotemporal semi-variogram

( ) ( )
Fig. 8  Empirical marginal temporal (left) T: 𝛾 0, hT and spatial (right) S: 𝛾 hS , 0 semi-variograms

( )
Finally, the pooled variogram 𝛾 h̃ where the spatial cor-
relation of the residual ϵ is considered constant over time
was also determined as follows:
12 N (hS ) �
� � 1 � � � � ��2 �
𝛾 h̃ = ∑12 � � 𝜖(s, t) − 𝜖 s + h S , t
2 i=1 N hS i=1 j=1
(11)

( ) i = 1, … , 12 correspond to the monitoring years and

where,
N hS are the number of pairs of points that are separated
by a distance h̃ in year t. The solid line in Fig. 9 is the fitted
model for the pooled variogram: a Gaussian model with a
nugget of 1.25, a range of 5725.92 m, and a sill of 5.77.

Fitting variogram models

Fig. 9  Pooled variogram

The parameters of the functions fitted to the marginals and

of non-stationarity in the time domain, as a consequence of pooled data semi-variograms were considered as initial
the continuous subsidence over time. values to fit the separable, product-sum, and sum-metric

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

381 Page 10 of 16 Environmental Earth Sciences (2022) 81:381

Fig. 10  Wireframe plots of empirical spatiotemporal semi-variogram of the regional subsidence in Mexico City lacustrine area between the
period 1983–2007(upper left corner) and fitted space–time variograms: separable, product-sum and sum-metric

Table 1  Root mean square Model RMSD To assess the goodness of the fit of the models and the
deviation empirical semi-variogram, the root-mean-square-deviation
Separable 0.0645 (RMSD) was determined between the surfaces (Table 1).
Product-sum 0.0663 Table 1 shows that the separable model characterizes
Sum-metric 0.0819 adequately the behavior of the data by having the lowest
value of the RMSD. The solid lines in Fig. 11 are the Gauss-
ian functions fitted to the marginal experimental semi-vario-
variogram models using the R gstat package (Eq. (4), (5) grams obtained from the separable model. This figure shows
and (7)). Figure 10 shows the empirical spatiotemporal semi- that this model represents adequately the variation in space
variogram of the phenomenon and the fitted models. In this and time of the phenomenon.
figure, surfaces colors represent the semivariance values After fitting a theoretical model to the empirical semi-
variation between 0 (purple) to 9 (yellow), approximately. variogram, predictions of the phenomenon in space and time
On the other hand, this figure shows that apparently, the are made by kriging.
separable and product-sum surfaces are alike, as the latter
is an extension of the previous one.

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2022) 81:381 Page 11 of 16 381

( ) ( )
Fig. 11  Empirical marginal temporal (left) T: 𝛾 0, hT and spatial (right) S: 𝛾 hS , 0 semi-variograms, and fitted variograms

Fig. 12  Ground elevation values (masl) between the period 2010–2030 at the points where the benchmarks are located

Predictions
To estimate the evolution of the regional subsidence phe-
nomenon in Mexico City, standard kriging was performed
up to the year 2030, considering the benchmark locations as
estimation points. The results obtained are shown in Fig. 12
where the red points correspond to the lowest ground eleva-
tions predicted for each year. For the year 2030, maximum
cumulative subsidence of 13.16 m was predicted in bench-
mark B(N01E06)02 near the Mexico City International
Airport.
Once the ground elevation values were determined for
each of the benchmarks for the period 2010–2030, the
mean subsidence rates were calculated in centimeters per
year. To compare these values with the ones determined
between 1983 and 2007, the same classes obtained from the
Jenks Natural Breaks method were used to organize them
(Fig. 4). Figure 13 shows that the highest subsidence rates
predicted during this period are between 23.1 and 28.8 cm/
year which corresponds to the cyan points located close to
the Mexico City International Airport. The highest mean
rate was 25.06 cm/year, which was recorded in benchmark
B(N01E06)02 where the maximum cumulative subsidence
for the year 2030 is presented.
To obtain subsidence rate contours lines a standard krig-
ing interpolation was performed on a 100 m × 100 m grid
considering a Gaussian variogram model with nugget 10,
sill 60, and range 6000 m (Fig. 14). Figure 15 shows a mean
subsidence rate map between 2007 and 2030 where red
zones correspond to areas where subsidence rates are higher,

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

381 Page 12 of 16 Environmental Earth Sciences (2022) 81:381

Fig. 13  Subsidence rate of each

superficial benchmark during
2007–2030

Comparing the subsidence rates determined between

Fig. 14  Semi-variogram of the predicted mean subsidence rate in the
period 2007–2030
1983 and 2007 with the ones predicted from the geostatisti-
cal model between 2007 and 2030, it can be observed that
they slow by approximately 30% and that is in agreement
with results obtained by other authors (Avilés and Pérez-
Rocha 2010), the largest changes in subsidence rate values
are associated with the thickness of the clay deposits. How-
ever, it is important to notice that this model simplifies a
complex process where the rate of groundwater extraction,
volumetric compressibility, and hydrogeological factors
between others, are involved, and that is not considered.
This model represents the spatial and temporal correlation
of ground benchmark readings located in the lacustrine zone
of Mexico City over 24 years of monitoring.
Validation of the predictions

The validation process consisted of comparing the direct

and the blue ones, where they are lower. Finally, Fig. 16 measurements made in the year 2016 in the benchmarks with
illustrates the behavior of the predicted ground elevation val- the ground elevations obtained from the model for the same
ues between 2007 and 2030 within subsidence rate classes. year. This process was made to verify that the predicted

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2022) 81:381
Fig. 15  Mean subsidence rate
map in the period 2007–2030
values from the space–time model represent physically the
behavior of regional subsidence in Mexico City.
Equation 12 was used to determine the error, ε, between
the registered ground elevations and those predicted by the
space–time model.
𝜀 = measured value − predicted value (12)
Figure 17 shows the box-plot of the errors in meters
where it can be observed that they have little variability
(𝜎 2 = 0.21), a mean equal to 0.08 m, and a few outliers.
In addition, Fig. 18 shows the spatial distribution of these
errors in absolute values. This figure shows that at 99% of
the sites the differences between the measured and predicted
value are less than 1.00 m. This implies that in general accu-
rate ground elevation values and subsidence rates can be
obtained from the proposed space–time model during the
period 2010–2030 for the lacustrine zone of Mexico City.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 13 of 16 381
Discussion and conclusions
As there is not a specific theoretical or empirical method
to analyze the regional subsidence in the lacustrine area of
Mexico City, a geostatistical method was developed to esti-
mate the spatial and temporal evolution of this phenomenon
between 2010 and 2030 by considering topographic refer-
ence information recorded between 1983 and 2007. How-
ever, it is important to notice that the model presented in
this paper simplifies a complex phenomenon where geologi-
cal, hydrogeological, and geotechnical factors are involved,
which were not included. Furthermore, limited spatial infor-
mation was considered because a full grid layout was used,
as a first approximation, to determine the spatial and tem-
poral correlation of the phenomenon in the area of study.
This grid represents a combination of spatial and temporal
attributes, where every spatial feature has a time observation
point (i.e., benchmarks with missing level readings in any
monitored year was not considered). An improvement of the
model could be made by considering a sparse grid layout to
include more spatial information.
13
381 Page 14 of 16 Environmental Earth Sciences (2022) 81:381

Fig. 16  Evolution of the ground elevations during the period 2007–2030 at the points where the benchmarks are located

The benchmarks classification by subsidence rate classes

was made to enhance the spatial variability of the subsidence
phenomenon in the lacustrine zone, and to visualize their
different trends over time.
The subsidence rate map between 1983 and 2007 obtained
from the interpretation of leveling data have an overall cor-
respondence in magnitude and spatial distribution of low and
high values with InSAR, PSI and SqueeSAR maps obtained
from earlier studies.
Based on the RMSD value, the separable model was
the one that best represented the spatial and temporal cor-
relation of the observations of the phenomenon in the area
of study, so it was considered to perform spatio-temporal
kriging to predict the ground level values during the period
Fig. 17  Box-plot of the errors in meters

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2022) 81:381
Fig. 18  Distribution of the absolute errors in meters
2010–2030. It is important to notice that this model assumes
that the temporal behavior of the process is the same for all
spatial points, i.e., this model does not allow changes in the
spatial pattern over time.
Although geotechnical factors such as the variability of
soil properties, stress conditions, and stress histories in clay
masses were not included in the geostatistical model, the
largest changes in subsidence values are associated with
the thickness of the clay deposits in the lacustrine zone of
Mexico City.
Predictions were made until 2030 because the results
obtained at the sites where the benchmarks are located
showed that the phenomenon would follow a constant behav-
ior implying that this consolidation process will eventually
stop, which is incorrect. As the time of prediction increases
the uncertainty of the results will also increase because of
the lack of recent level information that could be included in
the model. Although this geostatistical model has its limita-
tions, it provides valuable information that could help the
respective authorities to make better decisions about further
urban development and to identify the existing infrastructure
that could suffer important damages because of differential
settlement.
The validation process consisted of comparing the direct
measurements made in the year 2016 in the benchmarks with
the ground elevations obtained from the model for this same
year. It was observed that in 99% of location points, differ-
(ences were) less than 1.00 m and they had little variability
𝜎 2 = 0.21 around the mean (𝜇 = 0.08m) . Nevertheless,
another validation process could be made by comparing
the results with a digital elevation model available for the
region.
In the next stage of this work, the changes in domi-
nant periods of clay deposits caused by this phenomenon
would be assessed in the lacustrine zone of Mexico City
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 15 of 16 381
considering the ground measurements obtained from this
space–time model and several assumptions.
Funding The paper has not funding for the preparation.
Declarations
Conflict of interest We declare that the authors have no conflict of in-
terest.
References
Aguilera P (2013) Catedral Metropolitana: hundimiento y rescate.
Chapter II. Universidad Nacional Autónoma de México. Instituto
de Ingeniería, Mexico
Arroyo D, Ordaz M, Ovando-Shelley E, Guasch JC, Lermo J, Pérez
C, Alcántara L, Ramírez-Centeno MS (2013) Evaluation of the
change in dominant periods in the lake-bed zone of Mexico City
produced by ground subsidence through the use of site amplifica-
tion factors. Soil Dyn Earthq Eng 44:54–66
Avilés J, Pérez-Rocha L (2010) Regional subsidence of Mexico City
and its effects on seismic response. Soil Dyn Earthq Eng 44:54–66
Bard PY, Campillo M, Chávez-García FJ, Sanchez-Sesma F (1988) The
Mexico earthquake of September 19, 1985: a theoretical investiga-
tion of large-and small-scale amplification effects in the Mexico
City Valley. Earthq Spectra 4:609–633
Bilonick RA (1988) Monthly hydrogen ion deposition maps for the
northeastern U.S. from, July 1982 to September 1984. Atmos
Environ 22(9):1909–1924
Bivand RS, Pebesma E, Gómez-Rubio V (2013) Applied spatial data
analysis with R, 2nd edn. Springer, New York
Cabral-Cano E, Dixon TH, Miralles-Wilhelm F, Díaz-Molina O,
Sánchez-Zamora O, Carande RE (2008) Space geodetic imag-
ing of rapid land subsidence in Mexico City. Bull Geol Soc Am
120:1556–1566
Cabral-Cano E, Díaz Molina O, Delgado-Granados H (2011) Subsid-
encia y sus mapas de peligro: un ejemplo en el área nororiental
de la Zona Metropolitana de la Ciudad de México. Boletín De La
Soc Geol Mexicana 63:53–60
13
381 Page 16 of 16
Carrillo N (1948) Influence of artesian wells in the sinking of Mex-
ico City. Proceedings of the IIth Congress of International Soil
Mechanics and Foundation Engineering, Rotterdam 6:156–159
Chaussard E, Wdowinski S, Cabral-Cano E, Amelung F (2014) Land
subsidence in central Mexico detected by ALOS InSAR time-
series. Remote Sens Environ 140:94–106
Cressie NA (1993) Statistics for spatial data. Wiley, New York
Cruickshank C, Herrera I, Yates R, Hennart JP, Balarezo D, Magaña
del Toro R (1979) Modelo de predicción del hundimiento del
subsuelo del Valle de México. Instituto de Ingeniería, UNAM.
Elaborado para el Departamento del Distrito Federal, México.
De Cesare L, Myers D, Posa D (2001) Estimating and modeling space-
time correlation structures. Statist Probab Lett 51(1):9–14
De Iaco S, Myers D, Posa D (2001) Space-time analysis using a general
product-sum model. Statist Probab Lett 51(1):9–14
De Iaco S, Myers D, Posa D (2011) One strict positive definiteness of
product and product-sum covariance models. Journal of Statistical
Planning and Inference 141:1132–1140
González-Morán T, Rodríguez R, Cortes SA (1999) The Basin of
Mexico and its metropolitan area: water abstraction and related
environmental problems. J S Am Earth Sci 12:607–613
Gräler B, Pebesma E, Heuvelink G (2016) Spatio-Temporal Interpola-
tion using gstat. R J 8(1):204–2018
Hernández-Espriú A, Reyna-Gutiérrez JA, Sánchez-León E, Cabral-
Cano E, Carrera-Hernández J, Martínez-Santos P, Macías-
Medrano S, Falorni G, Colombo D (2014) The DRASTIC-Sg
Model: a new extension to the standard DRASTIC approach for
mapping groundwater vulnerability in urban aquifers subject to
differential land subsidence with application to Mexico City.
Hydrogeol J 22:1469–1485
Herrera I, Martínez R, Hernández G (1989) Contribución para la
administración científica del agua subterránea de la Cuenca de
México. Geofís Int 28:297–334
Jaime A (1988) Geotecnia y Sismicidad en el Valle de México. Reporte
No. D. Serie Azul, Instituto de Ingeniería, UNAM. Mexico.
Jenks GF, Caspall FC (1971) Error on choroplethic maps: definition,
measurement, reduction. Ann Am Geogr 61:217–244
Lesser JM, Cortés M (1998) El hundimiento del terreno en la ciudad de
México y sus implicaciones en el sistema de drenaje. Ingeniería
Hidráulica En México 3(13):13–18
Marsal M (1996) La isla de los perros. El Colegio Nacional, México
Marsal RJ, Mazari M (1959) The Subsoil of Mexico City. Escuela de
Ingeniería. Universidad Nacional Autónoma de México, Mexico
Marsal RJ, Graue R (1969) El subsuelo del lago de Texcoco. Nabor
Carrillo Conmemorative Volume: 167–202.
Mesri G, Rohsar B, Bohor BF (1975) Composition and compressibility
of typical samples of Mexico City clay. Geotechnique 25:527–554
Ortiz-Zamora D, Ortega-Guerrero A (2010) Evolution of long-term
land subsidence near Mexico City: Review, field investigations,
and predictive simulations. Water Resour Res 46:1–15
Osmanoǧlu B, Dixon TH, Wdowinski S, Cabral-Cano E, Jiang Y
(2011) Mexico City subsidence observed with persistent scatterer
InSAR. Int J Appl Earth Obs Geoinf 13:1–12
Ossa A, Romo MP (2007) The sinking of Mexico City: its effects
on soil properties and seismic response. Soil Dyn Earthq Eng
44:54–66
Ossa A, Botero E, Madrigal MC, Ovando E, Mendoza M, López-
Acosta NP (2019) Performance of a pavement foundation system
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Environmental Earth Sciences (2022) 81:381
based on the partial compensation of masses method. Soils Found
59:351–366
Ovando E, Romo MP, Contreras N, Giralt A (2003) Effects on soil
properties of future settlements in downtown Mexico City due to
ground water extraction. Geofísica Internacional 42(2):185–204
Ovando E, Ossa A, Romo MP (2007) The sinking of Mexico City: its
effects on soil properties and seismic response. Soil Dyn Earthq
Eng 27(4):333–343
Ovando E, Ossa A, Santoyo E (2013) Effects of regional subsidence
and earthquakes on arquitectural monuments in Mexico City.
Boletín De La Sociedad Geológica Mexicana 65(1):157–167
Pebesma E (2012) spacetime: spatio-temporal data in R. J Stat Softw
51:7
R Core Team (1993) R: a Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna
Romo MP, Jaime A, Reséndiz D (1988) Mexico earthquake of Sep-
tember 19, 1985: general soil conditions and clay properties in
the Valley of Mexico. Earthq Spectra 4:731–752
RStudio Team (2015) RStudio: Integrated Development for R. RStudio
Inc, Boston
SACMEX (2005) Programa de Gestión Integral de los Recursos Hídri-
cos. Sistema de Aguas de la Ciudad de México. Administración
Pública del Distrito Federal. Gobierno del Distrito Federal.
Décima Quinta Época. 27 de mayo de 2005. No.62-BIS.
Santoyo-Villa E, Ovando-Shelley E, Mooser F, León Plata E (2005)
Síntesis Geotécnica de la Cuenca del Valle de México. TGC,
México D.F.
Shapiro NM, Singh SK, Almora D, Ayala M (2001) Evidence of the
dominance of higher-mode surface waves in the lake-bed of the
Valley of Mexico. Geophys J Int 147:517–527
SIMOH (2015) Sistema de Monitoreo de la Piezometría y de los Hun-
dimientos del Valle de México por Extracción de Agua Subter-
ránea. Informe Final elaborado para la Comisión Nacional del
Agua (CONAGUA). Convenio: CGPEAYS-UNAM-04/2013.
Instituto de Ingeniería de la UNAM.
Snepvangers J, Heuvelink G, Huisman J (2003) Soil water content
interpolation using spatio—temporal kriging with external drift.
Geoderma 112:253–271
Tamez E, Ovando-Shelley E, Santoyo E (1997) Underexcavation of the
Metropolitan Catedral in Mexico City. Proceedings of the XIVth
International Conference on Soil Mechanics and Foundation Engi-
neering, Special Invited Lecture, Hamburg 4:2105–2126
Tamez E, Santoyo E, Cuevas A, Ovando-Shelley E (1995) Diagnóstico
y proyecto geotécnico. Chapter II. In: Catedral Metropolitana: cor-
rección geométrica, informe técnico. Mexico City: Asociación de
Amigos de la Catedral Metropolitana de México, A.C, Zaldívar,
S., 41–114.
Yin JH, Graham J (1994) Equivalent times and one-dimensional elastic
viscoplastic modelling of time-dependent stress-strain behavior of
clays. Can Geotech J 31(1):42–52
Yin JH, Graham J (1996) Modelling of one-dimensional consolidation.
Geotechnique 46(3):515–527
Publisher's Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

onlineservice@springernature.com