Repositorio Institucional (Univ. de la Frontera)

Esta investigación fue financiada por la Comisión Nacional de Investigación Científica y Tecnológica de Chile (CONICYT) a través del Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYT)- Proyecto 11170525

Cárcamo-Oyarzun, Jaime

Estevan, Isaac

Herrmann, Christian

Association between Actual and Perceived Motor Competence in School Children

Letelier, Viviana

Moreno Ortega, Jose

Maria Tremino, Rosa

Henriquez-Jara, Bastian, I

Fustos, Ivo

Real-Herraiz, Teresa

Moriconi, Giacomo

Angel Climent, Miguel

Sanchez, Isidro

English

The Use of Volcanic Powder as a Cement Replacement for the Development of Sustainable Mortars

Delgado, M.

Zuniga-Feest, A.

Reyes-Diaz, M.

Barra, P. J.

Ruiz, S.

Bertin-Benavides, A.

Valle, S.

Pereira, M.

Lambers, H.

Ecophysiological Performance of Proteaceae Species From Southern South America Growing on Substrates Derived From Young Volcanic Materials

Montesdeoca, Fabian

Avila, Maria

Quishpe, Janeth

Borie, Fernando

Cornejo, Pablo

Aguilera, Paula

Alvarado, Soraya

Espinosa, Jose

EARLY CHANGES IN THE TRANSITION FROM CONVENTIONAL TO NO-TILLAGE IN A VOLCANIC SOIL CULTIVATED WITH BEANS (Phaseolus vulgaris L.)

Evaluating the Performance of Various Feeding Systems for Hens

Statistical Evaluation of Weight, Yolk, Albumen and Shell in Egg Production

Canario, Joao Paulo

de Mello, Rodrigo Fernandes

Curilem, Millaray

Huenupan, Ferno

Rios, Ricardo Araujo

Data in Brief 30 (2020) 105627 
Contents lists available at ScienceDirect 
Data in Brief 
journal homepage: www.elsevier.com/locate/dib 
Data Article 
Llaima volcano dataset: In-depth comparison 
of deep artificial neural network architectures 
on seismic events classification 
João Paulo Canário a , ∗, Rodrigo Fernandes de Mello b , 
Millaray Curilem c , Fernando Huenupan c , Ricardo Araujo Rios a 
a Department of Computer Science, Federal University of Bahia, Brazil 
b Institute of Mathematics and Computer Science, University of São Paulo, Brazil 
c Department of Electrical Engineering, Universidad de La Frontera, Chile 
a r t i c l e i n f o 
Article history: 
Received 14 April 2020 
Accepted 20 April 2020 
Available online 30 April 2020 
Keywords: 
Volcano Monitoring 
Time Series Analysis 
Time Series modelling 
Neural Network 
a b s t r a c t 
This data manuscript presents a set of signals collected from 
the Llaima volcano located at the western edge of the Andes 
in Araucania Region, Chile. The signals were recorded from 
the LAV station between 2010 and 2016. After individually 
processing and analyzing every signal, specialists from the 
Observatorio Vulcanológico de los Andes Sur (OVDAS) clas- 
sified them into four class according to their event source: i) 
Volcano-Tectonic (VT); ii) Long Period (LP); iii) Tremor (TR), 
and iv) Tectonic (TC). The dataset is composed of 3592 sig- 
nals separated by class and filtered to select the segment that 
contains the most representative part of the seismic event. 
This dataset is important to support researchers interested 
in studying seismic signals from active volcanoes and devel- 
oping new methods to model time-dependent data. In this 
sense, we have published the manuscript “In-Depth Com- 
parison of Deep Artificial Neural Network Architectures on 
Seismic Events Classification” [1] analyzing such signals with 
different Deep Neural Networks (DNN). The main contribu- 
tion of such manuscript is a new DNN architecture called 
SeismicNet, which provided classification results among the 
best in the literature without demanding explicit signal pre- 
∗ Corresponding author: João Paulo Canário Federal University of Bahia, Department of Computer Science, Av Ademar 
de Barros Avenue, 40170-110 Salvador, Brazil, Phone: + 5571999538382 
E-mail addresses: joao.canario@ufba.br , jopacanario@gmail.com (J.P. Canário). 
https://doi.org/10.1016/j.dib.2020.105627 
2352-3409/© 2020 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license. 
( http://creativecommons.org/licenses/by/4.0/ ) 
2 J.P. Canário, R.F. de Mello and M. Curilem et al. / Data in Brief 30 (2020) 105627 
processing steps. Therefore, the reader is referred to such 
manuscript for the interpretation of the data. 
© 2020 The Authors. Published by Elsevier Inc. 
This is an open access article under the CC BY license. 
( http://creativecommons.org/licenses/by/4.0/ ) 
V
 
1
 
i  
t  
s  
i  Specifications table 
Subject Environmental Science and Artificial Intelligence 
Specific subject area Application of neural networks in the modelling of seismic patterns 
Type of data 1D Time-Frequency Signal 
How data were 
acquired 
The signals produced by different seismic events were collected from the Llaima volcano 
located at the western edge of Andes in Araucania Region (S 38o41 ′ - W 71o44 ′ ), Chile. 
The state agency responsible for monitoring this volcano and processing its signals is the 
Observatorio Vulcanológico de los Andes Sur (OVDAS). 
Data format Raw signals 
Parameters for data 
collection 
The signals were collected from LAV, that is one of the seven seismic stations of Llaima, 
being recorded in terms of the Z-vertical component from 2010 to 2016, sampled at 
100Hz and filtered using a 10th-order Butterworth bandpass filter in range [1, 10] Hz in 
order to preserve the bandwidth that contains the range of interest. The Z-vertical 
component contains all the necessary information to classify the event types, according 
to the specialists from OVDAS. 
Description of data 
collection 
The dataset was created from signals collected by monitoring stations located at the Llaima 
volcano (Chile). In summary, the signals represent activities related to the following 
events: 
• Volcano-Tectonic (VT): events associated with brittle failure of rocks inside of the 
volcanic building, which is the same type of event that happens along purely tectonic 
faults (e.g. the San Andreas Fault); 
• Long Period (LP): events usually observed before volcanic eruptions, being an important 
indicator of relevant imminent activities; 
• Tremor (TR): events that produce continuous and high-amplitude signals originated 
from several different processes, such as long-lived resonance due to extended flow of 
magma movement through cracks; 
• Tectonic (TC): events are not related to volcanic activities, being a typical result of the 
dynamics of geological faults. 
Data source location Llaima Volcano located at the western edge of Andes in Araucania Region (S 38 o 41 ′ - W 
71 o 44 ′ ), Chile 
Data accessibility The data is available at http://dx.doi.org/10.17632/dv8nwdd36k.1 
Related research article Canário, J. P., de Mello, R. F., Curilem, M., Huenupan, F., Rios, R. A. (2020). In-Depth 
Comparison of Deep Artificial Neural Network Architectures on Seismic Events 
Classification. Journal of Volcanology and Geothermal Research. 
alue of the data 
• The data can inspire new methods to detect and classify signals. 
• Our data is helpful to understand different patterns from volcano seismic events. 
• The data is important to compare the behavior of similar events from different active volca-
noes. 
. Data description 
The signals were collected from the Llaima volcano located at the western edge of Andes
n Araucania Region (S 38 o 41 ′ - W 71 o 44 ′ ), Chile. Due to its location, Llaima is considered a
ouristic attraction surrounded by villages, whose productive activity is mainly farming and live-
tock. Aiming at providing some security level for the people living in the neighborhood, there
s a state agency, called OVDAS (Observatorio Vulcanológico de los Andes Sur), that monitors not
J.P. Canário, R.F. de Mello and M. Curilem et al. / Data in Brief 30 (2020) 105627 3 
Fig. 1. Llaima volcano and its seismic stations [3] . 
 
 
 
T  
 
 
 
 
 
 
only Llaima but also other 42 volcanoes over the whole country. In particular for Llaima, OVDAS
performs constant surveillance with 9 stations that continuously gather seismic activity with a
24/7 monitoring service as represented in Fig. 1 . 
In particular, the data presented here takes into account the signals under the types Volcano-
ectonic (VT), Long Period (LP), Tremor (TR) and Tectonic (TC) collected from LAV, one of the
seven seismic stations of Llaima, given it was the most complete dataset studied by the special-
ists from OVDAS, being recorded in terms of the Z-vertical component from 2010 to 2016, sam-
pled at 100Hz and filtered using a 10th-order Butterworth bandpass filter in range [ 1 , 10] Hz in
order to preserve the bandwidth that contains the range of interest. The Z-vertical component
contains all the necessary information to classify the event types, according to the specialists
from OVDAS [ 1 , 3 ]. As a preprocessing step, a total of 3592 signals were normalized using their
4 J.P. Canário, R.F. de Mello and M. Curilem et al. / Data in Brief 30 (2020) 105627 
m  
T
 
r  
t  
i  
T  
H  
i  
i  
a  
n  
e  
s  
fl  
a  
H  
t  
r  
e  
t  
T  
t  
a  
t
2
 
d  
D  
t  
n  
t  
m  
f  
p
 
[  
i  
o  
a  
w  
r  
t  
w  
t  
o
 
o  
o  
F  
i  aximal values and organized into four classes with the following amounts: VT - 304, LP - 1310,
R - 490, and TC - 1488. 
As mentioned in the Table of Specifications and discussed in [ 1 , 3 ], Volcano-Tectonic (VT)
efers to brittle failure of rocks inside of the volcanic building, which is the same type of event
hat happens along purely tectonic faults. VT events are also resultant of normal tectonic forces
nside volcanoes due to the stress caused by the movement of fluids into pre-existing cracks.
his sort of activity presents a frequency pattern with a broadband spectrum that may reach 10
z. Events under the type Long Period (LP) correspond to the resonating of magma and gases
nside volcanic conduits toward the surface, whose spectral pattern is narrower than VT and it
s mainly bounded in [0.5, 5] Hz. LP events are usually observed before volcanic eruptions, being
n important indicator of relevant imminent activities. However, its occurrence is also part of the
ormal background seismicity at many volcanoes. Non-magmatic processes may also produce LP
vents, such as the case of glacier movements. Tremor (TR) are continuous and high-amplitude
ignals produced by several different processes, such as long-lived resonance due to extended
ow of magma movement through cracks, the continuous incidence of other events such as VT
nd LP when closely spaced over time. Their broadband spectrum is usually in range [0.5, 3.0]
z, being slowly attenuated at the end of the event. At last, Tectonic (TC) events are not related
o volcanic activities, being a typical result of the dynamics of geological faults. TCs may be the
esult of local, regional or even distant activities in terms of the epicenter location. When the TC
vent is detected far from the epicenter, it has lower frequencies than nearby ones. According
o the proximity of the source, TC could be misclassified as LP or VT. The spectral content of
C is similar to VT, being characterized by an impulsive beginning and an exponential decay, on
he other hand it typically carries more energy which is made evident by analyzing the signal
mplitudes. Fig. 2 illustrates an example of evey seismic class available in our dataset along with
heir spectrogram to better describe the signal behavior. 
. Experimental design, materials, and methods 
As discussed in [ 1 , 2 ], these volcano signals were analyzed using different approaches: raw
ata, spectrograms, and Wavelets. Experiments presented in [2] were designed to analyze the
NN performance while discriminating seismic activities through two common network archi-
ectures: a 2D Convolutional Neural Networks (CNN) and the Long Short-Term Memory (LSTM)
etwork [2] . In manuscript [1] , we extended the analysis presented in [2] by comparing those
wo architectures against the Multilayer Perceptron (MLP), which would be the simplest and
ost common baseline to take into account. Finally, a new architecture was proposed in [1] , re-
erred to as SeismicNet, allowed to accurately classify signals without demanding explicit signal
re-processing steps though. 
As commonly performed in the literature of artificial neural networks, we analyzed in
1] multiple architecture settings in terms of their loss (or accuracy performance if we consider
ts complement) in order to estimate the best as possible configurations to address the problem
f interest. The obtained results were evaluated by using measures traditionally computed to
ssess supervised learning tasks. Firstly, the generalization capability of our predictive models
as studied by sampling the original dataset using a 10-fold cross-validation strategy. Then, the
esults were organized into one-vs-all contingency matrices containing the number of true posi-
ive (TP), true negative (TN), false positive (FP), and false negative (FN). Based on such matrices,
e calculated four measures: i) Accuracy; ii) Error; iii) Specificity; and iv) F1-score. In addition
o these indices, the Kappa coefficient was also used to measure the general agreement between
ur classification system and experts, emphasizing the results were not obtained by chance. 
According to the results presented in [1] , one may notice the adjusted models were capable
f modeling and learning from seismic signals, by providing high accuracies within the interval
f 96.2% and 98.10%. Moreover, the other classification measures (Recall, Specificity, Kappa, and
1-score) were greater than 90%, whereas the error for all models were, at most, 3.1%. It is also
mportant to emphasize the outstanding results obtained with SeismicNet, showing classification
J.P.
 C
a
n
á
rio
,
 R
.F.
 d
e
 M
ello
 a
n
d
 M
.
 C
u
rilem
 et
 a
l.
 /
 D
a
ta
 in
 B
rief
 3
0
 (2
0
2
0
)
 10
5
6
2
7
 
5
 
Fig. 2. Samples of Llaima signals to represent every seismic event class: (A) LP; (B) TR; (C) VT; and (D) TC. 
6 J.P. Canário, R.F. de Mello and M. Curilem et al. / Data in Brief 30 (2020) 105627 
r  
W
A
 
2  
–  
c  
r  
c  
n
C
 
t  
R
[  
[  
 
[  
 
ates greater than 90%, without transforming the signals to images (outputs from Fourier and
avelet transforms) before proceeding with the training phase. 
cknowledgments 
This work was supported by FAPESP (Sao Paulo Research Foundation) under grants
013/07375-0 and CAPES (Coordination for the Improvement of Higher Education Personnel
Brazilian federal government agency). We gratefully acknowledge OVDAS (Observatorio Vol-
anologico de los Andes del Sur) for providing the datasets and the support from NVIDIA Corpo-
ation with the donation of the Titan V GPU used for this research. Any opinions, findings, and
onclusions or recommendations expressed in this material are those of the authors and do not
ecessarily reflect the views of FAPESP, NVIDIA and CAPES. 
onflict of Interest 
The authors declare that they have no known competing financial interests or personal rela-
ionships which have, or could be perceived to have, influenced the work reported in this article.
eferences 
1] J.P. Canário , R.F. de Mello , M. Curilem , F. Huenupan , R.A. Rios , In-Depth Comparison of Deep Artificial Neural Network
Architectures on Seismic Events Classification, Journal of Volcanology and Geothermal Research (2020) . 
2] Curilem, M., Canário, J.P., Franco, L., Rios, R.A., Using CNN to classify spectrograms of seismic events from Llaima
Volcano (Chile). In 2018 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE, https://doi.org/
10.1109/IJCNN.2018.8489285 . 
3] M. Curilem, R.F. de Mello, F. Huenupan, C. San Martin, L. Franco, E. Hernández, R.A. Rios, Discriminating seismic
events of the Llaima volcano (Chile) based on spectrogram cross-correlations, Journal of Volcanology and Geothermal
Research 367 (2018) 63–78 https://doi.org/10.1016/j.jvolgeores.2018.10.023 . 


Llaima volcano dataset: In-depth comparison of deep artificial neural network architectures on seismic events classification

Tremino, Rosa Maria

Ortega, Jose Marcos

Microstructure and Durability Performance of Mortars with Volcanic Powder from Calbuco Volcano (Chile) after 4 Hardening Years

Not a member yet