Content uploaded by José Francisco León-Cruz
Author content
All content in this area was uploaded by José Francisco León-Cruz on Jan 10, 2025
Content may be subject to copyright.
Tornadic environments in Mexico
Jos´
e Francisco Le´
on-Cruz
*
Unidad Acad´
emica de Estudios Territoriales Yucat´
an, Instituto de Geografía, Universidad Nacional Aut´
onoma de M´
exico (UNAM), Calle 50 419B, 97000 M´
erida,
Yucat´
an, Mexico
ARTICLE INFO
Keywords:
Tornadic environments
ERA5
Proximity soundings
Severe weather
K-means clustering
Mexico
ABSTRACT
Tornadoes represent a signicant threat to society. In Mexico, these natural hazards are common, principally
from the end of spring until autumn, with a mean of around 45 events yearly (2013−2022). Although there is no
ofcial tornado database in Mexico with a proper tornado classication, it is known that supercell and non-
supercell tornadogenesis is possible in the country. In this context, the present investigation examines the en-
vironments under 298 conrmed and validated tornadoes formed in the Mexican territory. Such analysis was
made using the proximity-sounding approach with the ERA5 reanalysis dataset. In addition, using the k-means
clustering method, three Tornadic Environment Types were found, each with specic characteristics. The rst
type is the most common environment, documented throughout the year, particularly during summer and the
beginning of autumn. Intermediate instability conditions, without wind shear, and high humidity near the
surface characterize it. The second type is observed in high altitudes during the spring, with relatively dry
conditions and low unstable environments. The previous examples may relate to non-supercell tornadogenesis in
different geographical regions and seasons. In contrast, the third type can be associated with signicant tor-
nadoes, an environment rich in instability and wind shear, concentrated in the northern portions of Mexico
during spring. The ndings of this research provide insights into increasing understanding of tornadoes in
Mexico. Furthermore, it can be helpful to generate improvements in tornado forecasting at the national level,
offering a range of tornadic environment types under which these natural hazards can develop. The clustering
method results offer an alternative option for the classication of tornadoes in countries with little capacity for
the ofcial classication of these phenomena.
1. Introduction
Tornadoes are considered one of the most extreme manifestations of
severe convective storms. These natural phenomena have been docu-
mented worldwide (Goliger and Milford, 1998; Maas et al., 2024), but
the United States of America (U.S.) is where tornadoes are mostly re-
ported. In general terms, tornadoes are categorized into two types:
supercell and non-supercell. The rst tornado type produces the most
intense convective vortices and requires the presence of a deep and
persistent mesocyclone, characteristic of supercell storms (Davies-Jones,
2015). The second tornado type produces a variety of intense atmo-
spheric vortices (e.g., landspouts and waterspouts) more related to the
intensication of preexisting, shallow vertical eddies near the surface
(Agee and Jones, 2009; Wakimoto and Wilson, 1989). Despite signi-
cant advances in understanding these natural phenomena, relevant as-
pects of their formation are still unknown, especially in countries with
relatively low frequency of events.
In Mexico, tornado research is still under development. Previous
tornado climatologies show these phenomena are relatively common in
central and northern regions of the country (Le´
on-Cruz et al., 2022;
Medrano and Avenda˜
no García, 2013). Interestingly, a signicant per-
centage of tornadoes in Mexico has been classied as non-supercell type,
and some authors have noted the possible role of the complex terrain in
the generation of these types of vortices (Le´
on-Cruz et al., 2019; Mon-
terde et al., 2023). Although tornadogenesis in Mexico seems to be a
primary non-supercell type, several signicant tornadoes have been
recorded, principally in the northeastern, such as those that occurred in
Cd. Acu˜
na on May 25, 2015; and the Piedras Negras (Mexico) – Rosita
Valley (U.S.) on April 24, 2007 (Barrett et al., 2017; Marshall and Eblen,
2008), both in Coahuila state. Other historical (Fuentes, 2010), social
(Macías-Medrano, 2016), and anthropological (Gonz´
alez P´
erez, 2019)
aspects of tornadoes have also been studied in Mexico. However, to date,
no research has been related to tornadic environments in the national
context.
* Corresponding author.
E-mail address: jleon@geograa.unam.mx.
Contents lists available at ScienceDirect
Atmospheric Research
journal homepage: www.elsevier.com/locate/atmosres
https://doi.org/10.1016/j.atmosres.2025.107916
Received 26 September 2024; Received in revised form 16 December 2024; Accepted 5 January 2025
Atmospheric Research 315 (2025) 107916
Available online 7 January 2025
0169-8095/© 2025 The Author. Published by Elsevier B.V. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ).
Previous research worldwide shows that tornadic environment an-
alyses are helpful for discrimination between strong/weak tornado
events and the differentiation from seasonal environmental character-
istics related to tornadogenesis (Rodríguez and Bech, 2021). Other
similar analyses show that this approach is helpful for the differentiation
between tornadic storms and other extreme events, such as windstorms
(Shikhov et al., 2021). This approach has also permitted to identify that
some particular parameters commonly used in tornadic forecasting
require a recalibration based on regional environmental conditions
(Hanesiak et al., 2024). The analysis of tornadic environments also ex-
tends to evaluating their projected changes under climate change con-
ditions (Kawazoe et al., 2023), a crucial aspect of risk management.
Furthermore, the increased understanding of which tornadic and non-
tornadic storms formed can help decrease the warning time and iden-
tify the characteristics of convective modes related to tornadoes (Coffer
et al., 2020).
Novel investigations on severe weather environments have also
focused on implementing clustering algorithms to identify, for example,
thunderstorm types. The above is particularly important because the
differentiation of these extreme events is independent of seasonal cri-
terion or static thresholds for typical instability and kinematic param-
eters (Morgenstern et al., 2023). This approach also helps, in the case of
hailstorms, to identify the convective mode from which these kinds of
storms are formed, but it also poses challenges for modeling the present
frequency and the response of hail under climate change conditions
(Zhou et al., 2021). In the case of tornadoes, recent investigations pro-
posed a clustering analysis that crucial instability and kinematic pa-
rameters are strongly inuenced by stational characteristics that drive
particular well-developed synoptic systems (Hanesiak et al., 2024).
Different research has test diverse clustering classication methods,
demonstrating the potential of these algorithms to group phenomena
with similar characteristics.
In this context, the present research aims to increase the under-
standing of tornadoes in Mexico by presenting a rst analysis of tornadic
environments in the country. The analysis of tornadic environments uses
the ERA5 reanalysis dataset, and, in addition, using an unsupervised
learning algorithm, different tornado types are presented for the
Mexican context. The results from this research can help improve the
forecasting of this type of extreme phenomena at the regional level.
Additionally, the results propose a starting point for tornado research in
Mexico in the context of climate change, dening the current charac-
teristics of the formation environments. The structure of the manuscript
is as follows. In section two, the data and methods are explained. In
section three, the results are shown and discussed, rst based on the
spatiotemporal view and then using the characteristics from the clusters
identied. Finally, conclusions are given.
2. Data and methods
2.1. Tornado database
This study used a recent tornado climatology in Mexico (Le´
on-Cruz
et al., 2022) as a starting point, covering the 2013–2020 period. Addi-
tionally, this climatology was updated (2021 and 2022) to consider at
least ten years of data. This update’s data collection, validation, and
homogenization follow the criteria described by Le´
on-Cruz et al. (2022).
The data collection comes from ofcial (e.g., reports from the National
Weather Services) and non-ofcial sources (e.g., media and social net-
works). The non-ofcial reports were veried using the cross-check
principle as well as meteorological data such as satellite products. All
tornado reports had a condence level assigned. It is important to note
that the lack of an ofcial tornado database for Mexico limits the ana-
lyses to only the country’s recent period. This updated database contains
420 reports: 309 tornadoes and 111 waterspouts. It is important to
mention that less than 5 % of the total reports hold an ofcial classi-
cation based on the Fujita or Enhanced Fujita scale; therefore, an
analysis based on intensity was discarded.
In order to reduce uncertainty in the analyses, diverse quality control
lters were applied. First, only tornadoes over land were considered, i.
e., waterspouts were excluded. The reason for this was the lack of reli-
able information on the location of the events over the sea since refer-
ence points for the georeferencing were lacking in most cases. Then,
tornado events without conrmation hours were also removed. Those
reports containing a period as a time of registration (i.e., early morning
00:00–05:59 LT, morning 06:00–11:59 LT, afternoon 12:00–17:59 LT, or
night 18:00–23:59 LT) were considered after an in-depth analysis of
each of these cases. Finally, all reports were inspected manually to
detect inconsistencies in essential elds such as location, date, and time.
In the end, 298 tornado reports (96.4 % of the original database) were
used for the analysis (Fig. 1).
The spatial pattern of tornado reports shows a concentration over the
central portions of the country, along with an area known as the Trans-
Mexican Volcanic Belt (Fig. 1). High activity also highlights the north-
western (including the North American Monsoon area) and northeastern
areas, where supercell storms are commonly reported (Weiss and Zeitler,
2008). Tornado season in Mexico covers from April to September, with a
rst peak in May and a second during July. Previous research (Le´
on-Cruz
et al., 2022) has mentioned the possible relationship of this behavior
with the passing of late cold fronts and the warm and moist season in
Mexico, which includes the presence of tropical cyclones and easterly
waves. About the diurnal cycle, tornadoes are more common during the
afternoon (12:00–18:00 local time) and the evening (18:00–00:00 local
time).
2.2. Reanalysis data and proximity soundings
In this research, the proximity sounding approach using the ERA5
reanalysis dataset was applied. This methodology considers the nearest
time (prior the tornado report) and position sounding for each event. For
this purpose, the ERA5 reanalysis dataset (Hersbach et al., 2020) was
selected. This dataset has a spatial resolution of 0.25 ×0.25 degrees, 1-h
time step, 37 pressure levels (1000–1 hPa), and spans from 1940 to the
present (Hersbach et al., 2023). Several research works have been
conducted using this reanalysis dataset to analyze tornadic, hailstorms,
and thunderstorm environments (Morgenstern et al., 2023; Rodríguez
and Bech, 2021; Zhou et al., 2021), which demonstrates its feasibility for
this kind of analyses. Other investigations also demonstrate the
acceptable performance of this database compared to rawinsonde ob-
servations (Li et al., 2020; Taszarek et al., 2021b).
Then, using the tornado database previously presented, a series of
vertical proles for each tornado case were produced using the NCAR
Command Language (NCL). A single vertical prole was generated for
tornado reports with conrmation hours. On the other hand, six proles
were generated for reports with a period as a time of registration, each
corresponding to a specic hour range (please see Section 2.1 for more
details). Then, in these cases, just one was retained based on CAPE
values: the most-unstable prole. The vertical proles contain the var-
iables of pressure level (hPa), altitude (m), temperature (◦C), dew point
temperature (◦C), wind direction (degrees), and wind speed (knots) in
the 37 pressure levels (1000–1 hPa) provided by the ERA5 reanalysis. It
is important to note that the use of these pressure levels instead of the
137 model levels decreases the availability of values near the surface;
therefore, only parameters between 2 and 3 km above the surface were
used. Given the complex topography in Mexico, all values for each
sounding under terrain elevation for each tornado report were removed.
Finally, the vertical proles generated were used to compute diverse
important parameters for tornado analysis, using the thundeR package
(Czernecki et al., 2024). This freeware R package allows computation of
over 200 thermodynamic, kinematic, and composite parameters that are
helpful in the analysis of severe storms and convective threats. The data
compilation of more than 200 indices for the 298 tornado cases was
integrated into the R software for statistical analysis and clustering.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
2
2.3. Clustering analysis
As previously mentioned, an aggrupation based on signicant and
non-signicant tornadoes was not possible given the low proportion
(less than 5 %) of cases with the proper classication on the Fujita or the
Enhanced Fujita scale. For this reason, a clustering analysis was imple-
mented to identify different Tornadic Environment Types (TET) in
Mexico. For this purpose, the K-means clustering method was selected.
This unsupervised machine learning algorithm was proposed in the late
‘60s (MacQueen, 1967) and aims to minimize the total variation within
each group. Recent investigations (Brown et al., 2023; Morgenstern
et al., 2023; Pacey et al., 2021) have used this method to identify
thunderstorm environment types.
For clustering analysis, the rst step was the variables selection.
Using previous research (Coffer et al., 2020; Hanesiak et al., 2024;
Kawazoe et al., 2023; Rodríguez and Bech, 2021), 12 variables useful for
the identication of tornadic environments were selected (Table 1).
These include instability, kinematic, and composite parameters. Then,
as a standard procedure prior to the application of the K-means algo-
rithm, the 12 parameters were standardized using z-scores.
The convective parameters include CAPE, computed in the most
unstable layer, which is a vertical integration that allows the identi-
cation of the instability available in the atmosphere, and CIN, also
computed in the most unstable layer, a parameter that allows mea-
surement the negative buoyant energy that suppresses convection.
Likewise, the LCL was included to know the altitude at which water
vapor condenses (as an approach of the cloud base height), and two
lapse-rates (between 2 and 4 km and 3–6 km above ground level), to
obtain information about the instability of the environment. The
moisture was also considered by including low-level relative humidity
(0–2 km), which is essential to convective initialization and thunder-
storm development, and 0–2 km moisture ux, which considered the
wind speed and mixing ratio in low levels.
On the other hand, the kinematic parameters considered were the
mid-level bulk shear (0–3 km) and deep-layer bulk shear (0–6), typically
used for discrimination between thunderstorms and severe thunder-
storms because of their importance in convective organization but also
their potential for mesocyclone generation. In the same vein, the SRH at
0–3 km above ground level was chosen. These parameters are useful to
measure the potential cyclonic updraft in thunderstorms, i.e., the po-
tential formation of right-moving supercells. Finally, two composite
Fig. 1. Spatial and temporal distribution of the tornado reports (2013−2023) used in this study. The dashed lines indicate three tornado hotspots: the Northwestern
and the Northeastern Mexico, and the Transmexican Volcanic Belt regions.
Table 1
Selected parameters for k-mean clustering.
Name Parameter Units
CAPE*Convective Available Potential Energy J kg
−1
CIN*Convective Inhibition J kg
−1
LCL*Lifting Condensation Level m
LR24 2–4 km lapse rate ◦C km
−1
LR36 3–6 km lapse rate ◦C km
−1
RH02 0–2 km relative humidity %
MF02 0–2 km moisture ux g s
−1
m
−2
BS03 0–3 km bulk shear m s
−1
BS06 0–6 km bulk shear m s
−1
SRH3 0–3 km storm-relative helicity m
2
s
−2
WMAX*WMAXSHEAR m
2
s
−2
EHI 0–1 km energy helicity index m
2
s
−2
*
CAPE, CIN, LCL, and WMAXSHEAR were calculated using the most-unstable
(MU) layer.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
3
parameters were computed: the rst, EHI (Rasmussen, 2003) 0–3 km
above ground level, considering SRH 0–3 km and surface-based CAPE;
and WMAXSHEAR, computed using the most-unstable layer, which used
CAPE and deep-layer bulk shear. Both parameters were used to
discriminate between possible supercells and non-supercell
environments.
The optimal number of clusters for K-means was selected by applying
the silhouette (Rousseeuw, 1987) and elbow (Thorndike, 1953)
methods, which resulted in two and three clusters, respectively (suppl.
Mat., S1). In addition, 26 different indices were computed using the
‘NbClust’ R package to determine the optimal number of clusters
(Charrad et al., 2014).The selected parameters for this package were a
Euclidean distance, a minimum and maximum number of clusters of 1 to
10, and the ‘K-means’ method. Following the majority rule, three clus-
ters were also identied as optimal (suppl. Mat., S2). In the end, the k-
means algorithm (R Core Team, 2020) was applied using a maximum
number of 500 iterations. In the next section, the results were discussed
based on the characteristics of the three clusters found.
3. Results and discussion
3.1. Spatial and temporal distribution
The Tornadic Environment Type 1 (TET1) considers 84 events
(28.19 %) majorly distributed in central and northern Mexico (Fig. 2).
The tornadoes classied with this TET are mostly documented during
the spring, from March to May, with an activity peak in April; however,
they are present during all the annual cycle (except for November and
December). The peak of activity during April coincides with the end of
the high activity of cold fronts in Mexico, according to the 1991–2020
climatology provided by the Mexican National Weather Service (SMN,
2024). The average distance to the coast for the tornadoes classied with
the TET1 is 263 km (the largest average compared to the other two
TETs). The average altitude of the TET1 is 1714 m asl; therefore, these
tornadoes are represented by those that occurred in high plains inland,
mainly during the spring.
Tornadoes classied with the TET2 consist of 175 events (58.72 %),
which represent the most common tornadic environment in Mexico
(Fig. 2). This TET is documented throughout the annual cycle but
principally during the end of spring until the beginning of autumn (from
May to September). In general terms, this tornado environment follows a
temporal distribution similar to the seasonal precipitation behavior in
Mexico’s central and southern portions (Colorado-Ruiz and Cavazos,
2021). Spatially, these kinds of events are present in practically the
entire national territory. The average distance to the coast of these
tornado reports is 160 km (the nearest distance compared to the other
two TETs); therefore, the role of systems such as tropical cyclones and
early waves (Agustín Bre˜
na-Naranjo et al., 2015; Dominguez et al.,
2020), also recorded during this period (more during summer and
autumn) seems crucial. The average altitude is 1543 m asl, so this TET2
can be summarized by those tornadoes that occurred near moisture
sources but in intermediate-high altitudes and during the warm and
rainy season in Mexico.
Finally, the TET3 comprises 39 tornado events (13.09 %). Their
spatial distribution is mainly along the north coast of the Pacic Ocean
(in the inuence area of the North American Monsoon) and the northern
portion of the Gulf of Mexico (Fig. 2). The tornadoes classied with this
TET are spatially located in the two regions previously identied with a
high potential of severe thunderstorms (Le´
on-Cruz et al., 2022), but also
in a region (in the northeast) where supercell storms have been docu-
mented (Weiss and Zeitler, 2008). These regions also coincide with those
where the frontal systems have greater inuence in Mexico. This TET is
more common during April and May (23/39 events), coinciding with the
end of spring. The average distance to the coast on the TET3 is 199 km.
However, the most interesting feature related to static conditions is that
the average altitude is only 798 m asl. In summary, these tornadoes
occur mainly over low plains where ocean moisture uxes seem relevant
and cold fronts (and the associated wind shear) are common.
Fig. 2. Spatiotemporal distribution of tornadoes in Mexico based on the Tornadic Environment Type (TET) obtained from clustering analysis. The upper panel shows
the number of reports (N) in each TET.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
4
3.2. Characteristics of the tornadic environment types
Based on CAPE (Fig. 3a), tornadoes in Mexico exhibit a general
median of 1141 J kg-1 and a 25th and 75th percentiles of 527 J kg
−1
and
1997 J kg
−1
, which is, in general, less than tornadoes reported in the U.
S. (Coffer et al., 2020), but similar to those documented in Spain
(Rodríguez and Bech, 2021) and Canada (Hanesiak et al., 2024), and
higher than those reported in Japan (Kawazoe et al., 2023). Based on the
environment types, the TET1 shows the lower median values (483 J
kg
−1
), which can be dened as low-instability environments; the TET2 is
considered an intermediate scenario with median values of 1360 J kg
−1
.
The TET3 is the most unstable environment (a median of 2759 J kg
−1
)
but also has higher variability.
The CIN parameter (Fig. 3b) shows a general median of −7 J kg
−1
,
and the medians computed for the TET1, TET2, and TET3 are −23 J
kg
−1
, −4 J kg
−1
, and −12 J kg
−1
, respectively. Only a few cases (mainly
in the TET1) show CIN values lesser than −75 J kg
−1
, which can be
considered a threshold for severe thunderstorms (Gensini and Ashley,
2011). In these cases, an additional mechanism for storm initialization
could have been required.
The LCL parameter (Fig. 3c) shows a general median of 850 m, as
well as the 25th and 75th percentiles of 500 m and 1300 m. The TET1
shows the highest LCL values, with a median of 1765 m, related to the
altitude at which tornadoes were generated (above 1700 m asl). The
median values for TET2 and TET3 are 650 m and 720 m, respectively,
concordant with the higher CAPE values. The computed LCL values for
the TET2 and TET3 are similar to those reported in the U.S., Canada, and
Japan (Coffer et al., 2020; Hanesiak et al., 2024; Kawazoe et al., 2023),
for example, but for the case of TET1, the reported values are over the
median commonly reported for other regions.
Continuing with the stability evaluation, the lapse rates at different
levels (Fig. 4a and b) generally show a median value of −6.1 ◦C for 2–4
km and 3–6 km. Such computed values are very similar to those
observed in North America (Hanesiak et al., 2024; Sessa and Trapp,
2023), particularly for non-signicant tornadoes. The median distribu-
tions show very similar results according to the lapse rate for the
different TETs. In this sense, for the TET1 medians of −6.8 ◦C (LR 2–4
km) and −6.6 ◦C (LR 3–6 km) were found, while for the TET2 the
medians computed are −5.8 ◦C (LR 2–4 km) and −5.9 ◦C (LR 3–6 km)
and for TET3 the medians are −6.9 ◦C (LR 2–4 km) and −6.7 ◦C (LR 3–6
km). In the case of tornadoes classied as TET3, the values reported are
related to signicant tornadoes (EF2-EF5) in the U.S. (Sessa and Trapp,
2023) and those tornadoes reported mainly for the western portion of
Canada (Hanesiak et al., 2024). In most cases, the lower lapse rate
computed can be related to conditional instability, which, for the cases
reported here, is complemented by the presence of moisture.
Low-level moisture is considered a key element for tornadogenesis,
favoring the increase in latent instability. In this sense, two parameters
were included in this analysis: relative humidity from 0 to 2 km and
moisture ux from 0 to 2 km. The results from low-level relative hu-
midity (Fig. 4c) indicate that, generally, the median value is 72 %, and
the 25th and 75th percentiles range from 60 % to 81 %. Disaggregating
these values by TETs, the most humid environment is the TET2 (median
of 78 %), followed by the TET3 (median of 72 %) and, nally, the TET1
with a relatively dry environment with a relative humidity median of 53
%. These low values identied for the TET1 can be related to the higher
altitudes where these kinds of tornadoes are documented (above the
1700 m asl) and inland, where the ocean’s moisture is diluted after
passing through mountainous systems. The moisture ux computed
(Fig. 4d) reveals the inuence of moisture advection from the Gulf of
Mexico and the role of the North American Monsoon in the TET3 (52.3 g
s
−1
m
−2
). For the other two tornadic environments, the medians are
24.4 g s
−1
m
−2
(TET1) and 27.1 g s
−1
m
−2
(TET2). These values are
lower than those reported in other parts of the world, such as Canada
(Hanesiak et al., 2024).
For kinematic parameters, bulk shear and storm-relative helicity at
different layers were evaluated. Generally, 0–3 km BS is commonly used
for assessing the potential for tornadoes, while 0–6 km BS is used for
severe thunderstorms. Some research has proposed a threshold value of
12.5 m s
−1
in the 0–6 BS (Taszarek et al., 2021a) for determining severe
thunderstorm environments. However, such threshold values can vary
signicantly spatially. For the tornadic environments in Mexico, the
median value considering all tornado events is 5.7 m s
−1
(0–3 km BS)
and 7.5 m s
−1
(0–6 km BS), while the 25th and 75th percentiles are 3.5
m s
−1
and 8.3 m s
−1
for the 0–3 km BS and 4.6 m s
−1
and 13.0 m s
−1
for
the 0–6 km BS (Fig. 5a, b). Disaggregating values per tornadic envi-
ronments, only in the TET3, the bulk shear shows values similar to those
reported in the literature for tornadoes. In this sense, the median values
for 0–3 km BS are 13.0 m s
−1
, while for 0–6 km BS, the median is 16.1 m
s
−1
. These values are lower than those reported for the US (Coffer et al.,
2020; Sessa and Trapp, 2023) and Japan (Kawazoe et al., 2023) but
similar to those in central and western Canada (Hanesiak et al., 2024)
and Spain (Rodríguez and Bech, 2021).
A quite similar scenario is reported for the storm-relative helicity
parameter (Fig. 5c, d), where only in the TET3 are the median values
representative of signicant severe environments. In these types of
tornadoes (TET3), median value of 117 m
2
s
−2
is registered for 0–3 km
layer. In the rest of the TETs, the SRH does not seem crucial, with
Fig. 3. (a) Convective available potential energy, (b) convective inhibition, and (c) lifting condensation level boxplots disaggregated by Tornadic Environment Type
(TET). The dotted black line represents the median, while the grey shadow indicates the 25th and 75th percentiles for all tornadoes (i.e., without clustering). Values
were computed from proximity soundings using the ERA5 reanalysis dataset.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
5
median values under 32 m
2
s
−2
. In summary, kinematic parameters
seem important only for those tornadoes reported in northeastern and
northwestern Mexico (i.e., for the TET3), where signicant tornadoes
occur (Barrett et al., 2017; Le´
on-Cruz et al., 2017) and potential for
severe convective storms have been previously reported (Le´
on-Cruz
et al., 2022; Li et al., 2020).
Additionally, two additional composite were included in this anal-
ysis: the WMAXSHEAR (Taszarek et al., 2017) and the Energy Helicity
Index (Fig. 6). The WMAXSHEAR parameter shows a median value of
337 m
2
s
−2
for tornadoes without any clustering process, and the 25th
and 75th percentiles range from 186 m
2
s
−2
to 559 m
2
s
−2
. For the TET1
and TET2 the medians are 295 m
2
s
−2
and 289 m
2
s
−2
, respectively
Fig. 4. Same as Fig. 3 but for (a) 2–4 km lapse rate, (b) 3–6 km lapse rate, (c) 0–1 km relative humidity, and (d) 0–2 km moisture.
Fig. 5. Same as Fig. 3 but for (a) 0–3 km bulk shear, (b) 0–6 km bulk shear, (c) 0–3 km storm relative helicity, and (d) 0–3 km storm relative helicity.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
6
(Fig. 6a). In the case of the TET3, the median value is 1086 m
2
s
−2
,
signicantly larger than the general behavior and the other two tornadic
environments. It is important to mention that previous research about
tornadoes has shown that the WMAXSHEAR values associated with
these phenomena range from 500 to 1000 m
2
s
−2
(Chernokulsky et al.,
2023; Rodríguez and Bech, 2021).
A similar scenario is observed for EHI (Fig. 6b), but with median
values (generally and for TET1 and TET2) around zero. Some outliers
indicate values equal to and greater than 1 (for TET3), not too high to
indicate possible right-moving supercells. In the case of TET3, the me-
dian value is 1.47, with some outliers greater than 3. A similar distri-
bution has been observed for tornadoes documented in some portions of
Europe (Shikhov et al., 2021) but is higher than those reported for EF2+
tornadoes in Japan (Kawazoe et al., 2023), for example. These ndings
support the idea that tornadoes classied with the TET3 are likely
related to right-moving supercells, as previous works in the country have
identied (Barrett et al., 2017; Weiss and Zeitler, 2008). An in-depth
analysis of these events can be applied in the future, for example,
using numerical modeling or weather radar data.
In addition to the previous analysis by tornadic environments and
composite parameters, two other combinations between CAPE and deep-
layer (0–6 km) bulk shear were computed. The rst consider CAPE ×
BS06 ≥20,000 (Gensini and Ashley, 2011), and the second CAPE ×
BS061.6≥46,800 (Brooks et al., 2003). In the case of the WMAXSHEAR
(Taszarek et al., 2017), despite the lack of consensus on the threshold
value, a value of 500 m
2
s
−2
was selected as an approach previously used
for detecting thunderstorms and hailstorms in Mexico (Le´
on-Cruz et al.,
2023). For the three cases mentioned above, we compute the percentage
of events per TET exceeding the dened threshold values (Fig. 7). It is
important to mention that an additional threshold value of −75 J kg
−1
for CIN was considered for all the cases shown.
In the three CAPE and bulk shear combinations, the TET3 shows that
more than 84 % exceed such threshold values, indicating the high po-
tential for severe convective thunderstorms. For the tornadoes classied
with TET2, the percentage of cases exceeding the dened threshold
ranges from 9.7 % to 24.5 %, while for TET1, such values range from 3.5
% to 28.5 %. Interestingly, a similar percentage of cases exceeding the
threshold values (for all the TETs) is observed in the combination
Fig. 6. Same as Fig. 3 but for (a) WMAXSHEAR and (b) 0–3 km energy helicity index.
Fig. 7. Convective available potential energy versus 0–6km bulk shear. In (a), colored dots indicate proximity soundings where CAPE ×BS06 ≥20,000 (Gensini and
Ashley, 2011), in (b) where CAPE ×BS061.6≥46,800 (Brooks et al., 2003), and in (c) where
2CAPE
√×BS06 ≥500 (Taszarek et al., 2017). The percentage of
proximity soundings exceeding the same threshold values per tornadic environment type (TET) is also shown.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
7
proposed by Brooks et al. (2003) and Taszarek et al. (2017). The com-
bination proposed by Gensini and Ashley (2011) is the most conserva-
tive, with a lesser percentage of cases exceeding the threshold value.
4. Conclusions
This research investigates tornadic environments in Mexico using the
ERA5 reanalysis dataset and the proximity soundings approach. In
addition to analyzing the general characteristics of diverse instability
and kinematic parameters crucial for severe thunderstorms and tornado
development, a clustering analysis based on the k-means clustering
method was performed. In this sense, three different Tornadic Envi-
ronment Types (TET) were identied in Mexico, each with particular
characteristics.
TET1 is usually found in continental areas (they have the most
considerable mean distance from the oceans), in the central and north-
ern regions, and mainly during spring. The tornadoes classied with this
TET hold the least unstable conditions and the highest LCL, associated
with the higher altitudes where they are documented. Also, the TET1 is
characterized by relatively dry conditions without signicant moisture
convergence environments. The wind shear does not seem especially
crucial in this environment type, and the same goes for helicity. The
composite parameters computed do not show any indication of severity.
In summary, the TET1 could be related to less severe tornado events (for
example, non-supercell tornadogenesis) during the transition to late
winter-spring in Mexico, at higher altitudes with relatively dry
conditions.
TET2 is represented by the typical tornadoes related to thunder-
storms from the end of spring to summer and the beginning of autumn.
This type of environment occurs practically over the entire Mexican
territory, with emphasis on central and coastal regions. Tornadoes
grouped in TET2 show intermediate instability environments (based on
CAPE values over 1000 J kg
−1
) and low lifted condensation levels. Also,
this TET shows the lowest lapse rates but high-humidity environments
near the surface. Again, the wind shear does not seem relevant, and a
similar situation is observed for the storm-relative helicity. Composite
parameters, as expected, do not show any indication of severity. In
summary, the tornadoes classied as TET2 can be associated with non-
supercell phenomena in Mexico’s rainy and stormy season, principally
over intermediate and high-altitude regions relatively near the coasts. It
is important to note that this is Mexico’s most common tornadic
environment.
Finally, TET3 is represented by the most severe environment,
possibly with signicant tornadoes (EF2+) occurrence. These tornadoes
occur mainly at the end of spring in northeastern and northwestern
Mexico (in the inuence area of the North American Monsoon and the
Gulf of Mexico). This TET shows a very high instability environment
(based on CAPE values over 2500 J kg
−1
), a low lifted condensation
level, intense bulk shear and storm-relative helicity, high humidity in
low levels, and moisture convergence, representing ingredients for
supercell tornadogenesis. The composite parameters also show values
documented for signicant tornadoes (except for the SRH 0–3 km). For
the location of these reports, it is possible the inuence of moisture
uxes from the oceans (the Gulf of Mexico and the Pacic Ocean),
combined with a jet stream (for example, by the mountains ranges in the
region) that provides a shear environment, as well as the passage of cold
fronts.
Generally, tornadoes in Mexico developed in conditions similar to
other regions in the world (principally Europe) but less severe than in
the U.S. (Taszarek et al., 2020). The country’s geographic location
favored the presence of different atmospheric systems, such as tropical
cyclones and easterly waves, which provide humid conditions for storm
development. Other systems, such as cold fronts, also increase the po-
tential for frontal thunderstorms, which can also favor tornadogenesis.
The country’s complex terrain also can help moist orographic convec-
tion and set the conditions for thunderstorm development. Given that
most of the tornadoes in the country are non-supercell type, such
described conditions accompanied by a source of vorticity near the
surface can be enough for tornadogenesis.
This research provides insights into increasing understanding of
tornadoes in tropical and subtropical regions such as Mexico. Also, it
allows us to identify the different environments where tornadoes can
develop in the country, which is a rst step for generating improvements
in tornado forecasting. Implementing an unsupervised articial intelli-
gence algorithm, such as the K-means, provides a rst clue for the
possible classication of tornadoes in Mexico based on environmental
characteristics. Despite the limitations of the ERA5 reanalysis dataset,
for example, those related to the pressure levels available, or the
underestimating values reported by other authors, their use in countries
lacking state-of-the-art meteorological instrumentation (i.e., rawin-
sondes and weather radar data) is crucial to obtain information about
severe weather environments. Future studies should focus on imple-
menting additional datasets and methodologies (e.g., numerical
modeling or satellite products) to increase the understanding of these
natural hazards in the Mexican territory.
CRediT authorship contribution statement
Jos´
e Francisco Le´
on-Cruz: Writing – review & editing, Writing –
original draft, Visualization, Validation, Software, Methodology,
Investigation, Formal analysis, Data curation, Conceptualization.
Declaration of competing interest
The author declares no potential conict of interest.
Data availability
The data that support the ndings of this study are available on
request from the corresponding author.
Acknowledgements
The work of Karla Mu˜
noz Caballero and Arlen Gallegos Rodríguez
(undergraduate students) for updating the tornado database for the
period 2021-2022 is gratefully appreciated.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.atmosres.2025.107916.
References
Agee, E., Jones, E., 2009. Proposed conceptual taxonomy for proper identication and
classication of tornado events. Weather Forecast. 24 (2), 609–617. https://doi.org/
10.1175/2008WAF2222163.1.
Agustín Bre˜
na-Naranjo, J., Pedrozo-Acu˜
na, A., Pozos-Estrada, O., Jim´
enez-L´
opez, S.A.,
L´
opez-L´
opez, M.R., 2015. The contribution of tropical cyclones to rainfall in Mexico.
Phys. Chem. Earth Parts A/B/C 83-84, 111–122. https://doi.org/10.1016/j.
pce.2015.05.011.
Barrett, B.S., Farf´
an, L.M., Raga, G.B., Hern´
andez, D.H., 2017. The unusual early
morning tornado in Ciudad Acu˜
na, Coahuila, Mexico, on 25 May 2015. Mon.
Weather Rev. 145 (6), 2049–2069. https://doi.org/10.1175/MWR-D-16-0252.1.
Brooks, H.E., Lee, J.W., Craven, J.P., 2003. The spatial distribution of severe
thunderstorm and tornado environments from global reanalysis data. Atmos. Res.
67-68, 73–94. https://doi.org/10.1016/S0169-8095(03)00045-0.
Brown, A., Dowdy, A., Lane, T.P., Hitchcock, S., 2023. Types of severe convective wind
events in Eastern Australia. Mon. Weather Rev. 151 (2), 419–448. https://doi.org/
10.1175/MWR-D-22-0096.1.
Charrad, M., Ghazzali, N., Boiteau, V., Niknafs, A., 2014. NbClust: an R package for
determining the relevant number of clusters in a data set. J. Stat. Softw. 61, 1–36.
https://doi.org/10.18637/jss.v061.i06.
Chernokulsky, A., Shikhov, A., Yarinich, Y., Sprygin, A., 2023. An empirical relationship
among characteristics of severe convective storms, their cloud-top properties and
environmental parameters in Northern Eurasia. Atmosphere 14 (1), 174. https://doi.
org/10.3390/atmos14010174.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
8
Coffer, B.E., Taszarek, M., Parker, M.D., 2020. Near-ground wind proles of tornadic and
nontornadic environments in the United States and Europe from ERA5 reanalyses.
Weather Forecast. 35 (6), 2621–2638. https://doi.org/10.1175/WAF-D-20-0153.1.
Colorado-Ruiz, G., Cavazos, T., 2021. Trends of daily extreme and non-extreme rainfall
indices and intercomparison with different gridded data sets over Mexico and the
southern United States. Int. J. Climatol. 41 (11), 5406–5430. https://doi.org/
10.1002/joc.7225.
Czernecki, B., Taszarek, M., Szuster, P., 2024. Thunder: Computation and Visualisation
of Atmospheric Convective Parameters (Versi´
on 1.1.3) [R]. https://bczernecki.gith
ub.io/thundeR/.
Davies-Jones, R., 2015. A review of supercell and tornado dynamics. Atmos. Res. 158-
159, 274–291. https://doi.org/10.1016/j.atmosres.2014.04.007.
Dominguez, C., Done, J.M., Bruy`
ere, C.L., 2020. Easterly wave contributions to seasonal
rainfall over the tropical Americas in observations and a regional climate model.
Clim. Dyn. 54 (1), 191–209. https://doi.org/10.1007/s00382-019-04996-7.
Fuentes, O.V., 2010. The earliest documented tornado in the Americas: Tlatelolco,
August 1521. Bull. Am. Meteorol. Soc. 91 (11), 1515–1524. https://doi.org/
10.1175/2010BAMS2874.1.
Gensini, V.A., Ashley, W.S., 2011. Climatology of potentially severe convective
environments from the north American regional reanalysis. E-J. Severe Storms
Meteorol. 6 (8), 8. https://doi.org/10.55599/ejssm.v6i8.35.
Goliger, A.M., Milford, R.V., 1998. A review of worldwide occurrence of tornadoes.
J. Wind Eng. Ind. Aerodyn. 74-76, 111–121. https://doi.org/10.1016/S0167-6105
(98)00009-9.
Gonz´
alez P´
erez, D., 2019. Storm clouds, torrential storms and tornadoes: accounts of
waterspouts or tornado alleys in Oaxaca (1830-1885). Relac. Estudios Hist. Soc. 40
(159), 113–146. https://doi.org/10.24901/rehs.v40i159.631.
Hanesiak, J., Taszarek, M., Walker, D., Wang, C.-C., Betancourt, D., 2024. ERA5-based
signicant tornado environments in Canada between 1980 and 2020. J. Geophys.
Res. Atmos. 129 (8), e2023JD040614. https://doi.org/10.1029/2023JD040614.
Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Hor´
anyi, A., Mu˜
noz-Sabater, J.,
Nicolas, J., Peubey, C., Radu, R., Schepers, D., Simmons, A., Soci, C., Abdalla, S.,
Abellan, X., Balsamo, G., Bechtold, P., Biavati, G., Bidlot, J., Bonavita, M.,
Th´
epaut, J.-N., 2020. The ERA5 global reanalysis. Q. J. R. Meteorol. Soc. 146 (730),
1999–2049. https://doi.org/10.1002/qj.3803.
Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Hor´
anyi, A., Mu˜
noz Sabate, J., Nicolas,
J., Peubey, C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., &
Th´
epaut, J.-N. (2023). ERA5 Hourly Data on Pressure Levels from 1940 to Present.
[Dataset]. Copernicus Climate Change Service (C3S) Climate Data Store (CDS). Doi:
10.24381/cds.bd0915c6.
Kawazoe, S., Inatsu, M., Fujita, M., Sugimoto, S., Okada, Y., Watanabe, S., 2023.
Evaluation of tornadic environments and their trends and projected changes in
Japan. NPJ Clim. Atmos. Sci. 6 (1), 1–8. https://doi.org/10.1038/s41612-023-
00524-x.
Le´
on-Cruz, J.F., Carbajal, N., Pineda-Martínez, L.F., 2017. Meteorological analysis of the
tornado in Ciudad Acu˜
na, Coahuila State, Mexico, on May 25, 2015. Nat. Hazards 89
(1), 423–439. https://doi.org/10.1007/s11069-017-2972-6.
Le´
on-Cruz, J.F., Carbajal, N., Pineda-Martínez, L.F., 2019. The role of complex terrain in
the generation of tornadoes in the west of Mexico. Nat. Hazards 97 (1), 335–353.
https://doi.org/10.1007/s11069-019-03647-8.
Le´
on-Cruz, J.F., Pineda-Martínez, L.F., Carbajal, N., 2022. Tornado climatology and
potentially severe convective environments in Mexico. Clim. Res. 87, 147–165.
https://doi.org/10.3354/cr01692.
Le´
on-Cruz, J.F., Caetano, E., Cort´
es-Ramos, J., Dominguez, C., M´
endez-P´
erez, J.M., 2023.
Thunderstorm and hailstorm environments in Mexico. Int. J. Climatol. 43 (10),
4379–4395. https://doi.org/10.1002/joc.8093.
Li, F., Chavas, D.R., Reed, K.A., Ii, D.T.D., 2020. Climatology of Severe Local Storm
Environments and Synoptic-Scale Features over North America in ERA5 Reanalysis
and CAM6 Simulation. https://doi.org/10.1175/JCLI-D-19-0986.1.
Maas, M., Supinie, T., Berrington, A., Emmerson, S., Aidala, A., Gavan, M., 2024. The
Tornado Archive: Compiling and Visualizing a Worldwide, Digitized Tornado
Database. Bull. Am. Meteorol. Soc. 1 (aop). https://doi.org/10.1175/BAMS-D-23-
0123.1.
Macías-Medrano, J.M., 2016. Vulnerabilidad social en la Ciudad de M´
exico frente a
tornados. Rev. Mex. Sociol. 78 (2), 257–284.
MacQueen, J. (1967). Some methods for classication and analysis of multivariate
observations. Proc. Fifth Berkeley Symp. Math. Stat. Probabil., 1(14), 281–297.
https://books.google.com/books?hl=es&lr=&id=IC4Ku_7dBFUC&oi=fnd&pg
=PA281&dq=+Some+Methods+for+Classication+and+Analysis+of+Multivaria
te+Observations,&ots=nQUfI0NfvQ&sig=lH19M-ZEZ9w4jSoLE2oo7RrgbBI.
Marshall, T.P., Eblen, L., 2008. P3. 2 the Rosita Valley, TX Tornado: 24 April 2007. http
s://www.academia.edu/download/57181238/TIM-RositaValley.pdf.
Medrano, M.M., Avenda˜
no García, A., 2013. Climatología de tornados en M´
exico. Invest.
Geogr. 0 (83). https://doi.org/10.14350/rig.35726.
Monterde, D., Carbajal, N., Pineda-Martínez, L.F., Le´
on-Cruz, J.F., 2023. Analysis of a
multiple non-supercell tornado event in Mexico. Atmos. Res. 293, 106916. https://
doi.org/10.1016/j.atmosres.2023.106916.
Morgenstern, D., Stucke, I., Mayr, G.J., Zeileis, A., Simon, T., 2023. Thunderstorm
environments in Europe. Weath. Clim. Dyn. 4 (2), 489–509. https://doi.org/
10.5194/wcd-4-489-2023.
Pacey, G.P., Schultz, D.M., Garcia-Carreras, L., 2021. Severe convective windstorms in
Europe: climatology, preconvective environments, and convective mode. Weather
Forecast. 36 (1), 237–252. https://doi.org/10.1175/WAF-D-20-0075.1.
R Core Team, 2020. R A Language and Environment for Statistical Computing. R
Foundation for Statistical Computing. https://cir.nii.ac.jp/crid/1370298755
636824325.
Rasmussen, E.N., 2003. Rened supercell and tornado forecast parameters. Weather
Forecast. 18 (3), 530–535. https://doi.org/10.1175/1520-0434(2003)18<530:
RSATFP>2.0.CO;2.
Rodríguez, O., Bech, J., 2021. Tornadic environments in the Iberian Peninsula and the
Balearic Islands based on ERA5 reanalysis. Int. J. Climatol. 41 (S1), E1959–E1979.
https://doi.org/10.1002/joc.6825.
Rousseeuw, P.J., 1987. Silhouettes: a graphical aid to the interpretation and validation of
cluster analysis. J. Comput. Appl. Math. 20, 53–65.
Sessa, M.F., Trapp, R.J., 2023. Environmental and radar-derived predictors of tornado
intensity within ongoing convective storms. J. Oper. Meteorol. 11 (5). https://www.
researchgate.net/profile/Robert-Trapp-2/publication/370964885_Environment
al_and_Radar-Derived_Predictors_of_Tornado_Intensity_within_Ongoing_Convective
_Storms/links/64b53b88c41fb852dd7d6c2f/Environmental-and-Radar-Derived-Pre
dictors-of-Tornado-Intensity-within-Ongoing-Convective-Storms.pdf.
Shikhov, A., Chernokulsky, A., Kalinin, N., Bykov, A., Pischalnikova, E., 2021.
Climatology and Formation Environments of Severe Convective Windstorms and
Tornadoes in the Perm Region (Russia) in 1984–2020. Atmosphere 12 (11), 1407.
https://doi.org/10.3390/atmos12111407.
Servicio Meteorol´
ogico Nacional, S.M.N., 2024. Frentes Fríos. https://smn.conagua.gob.
mx/es/climatologia/pronostico-climatico/frentes-frios.
Taszarek, M., Brooks, H.E., Czernecki, B., 2017. Sounding-derived parameters associated
with convective hazards in Europe. Mon. Weather Rev. 145 (4), 1511–1528.
Taszarek, M., Allen, J.T., Púˇ
cik, T., Hoogewind, K.A., Brooks, H.E., 2020. Severe
convective storms across Europe and the United States. Part II: ERA5 environments
associated with lightning, large hail, severe wind, and tornadoes. J. Clim. 33 (23),
10263–10286. https://doi.org/10.1175/JCLI-D-20-0346.1.
Taszarek, M., Allen, J.T., Marchio, M., Brooks, H.E., 2021a. Global climatology and
trends in convective environments from ERA5 and rawinsonde data. NPJ Clim.
Atmos. Sci. 4 (1), 1–11. https://doi.org/10.1038/s41612-021-00190-x.
Taszarek, M., Pilguj, N., Allen, J.T., Gensini, V., Brooks, H.E., Szuster, P., 2021b.
Comparison of convective parameters derived from ERA5 and MERRA-2 with
rawinsonde data over Europe and North America. https://doi.org/10.1175/JCLI-D-
20-0484.1.
Thorndike, R.L., 1953. Who belongs in the family? Psychometrika 18 (4), 267–276.
https://doi.org/10.1007/BF02289263.
Wakimoto, R.M., Wilson, J.W., 1989. Non-supercell Tornadoes. Mon. Weather Rev. 117
(6), 1113–1140. https://doi.org/10.1175/1520-0493(1989)117<1113:NST>2.0.
CO;2.
Weiss, J., Zeitler, J., 2008. Supercells of the Serranías del Burro. In: 24th Conf. On Severe
Local Storms. https://ams.confex.com/ams/24SLS/techprogram/paper_141558.htm
.
Zhou, Z., Zhang, Q., Allen, J.T., Ni, X., Ng, C.-P., 2021. How many types of severe
hailstorm environments are there globally? Geophys. Res. Lett. 48 (23),
e2021GL095485. https://doi.org/10.1029/2021GL095485.
J.F. Le´
on-Cruz
Atmospheric Research 315 (2025) 107916
9