Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних
Kilauea is one of the most active and well-known volcanoes in the world and most of our knowledge of volcanism originates from its research. During a long study of volcanoes, many different methods of forecasting their activity were proposed, from the seismological analysis to the statistical analys...
Saved in:
| Date: | 2019 |
|---|---|
| Main Author: | |
| Format: | Article |
| Language: | English |
| Published: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2019
|
| Subjects: | |
| Online Access: | https://journal.iasa.kpi.ua/article/view/181465 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | System research and information technologies |
| Download file: |  |
Institution
System research and information technologies| _version_ | 1867334398731354112 |
|---|---|
| author | Zabielin, Stanislav |
| author_facet | Zabielin, Stanislav |
| author_institution_txt_mv | [
{
"author": "Stanislav Zabielin",
"institution": "Educational and Scientific Complex \"Institute for Applied System Analysis\" of the National Technical University of Ukraine \"Igor Sikorsky Kyiv Polytechnic Institute\", Kyiv"
}
] |
| author_sort | Zabielin, Stanislav |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2020-03-02T17:05:10Z |
| description | Kilauea is one of the most active and well-known volcanoes in the world and most of our knowledge of volcanism originates from its research. During a long study of volcanoes, many different methods of forecasting their activity were proposed, from the seismological analysis to the statistical analysis of their emissions. However, a comprehensive analysis of data arrays with the help of intelligent methods of data analysis has not been carried out before. Using fuzzy data processing methods, a neural network, volcanic and atmospheric indicators, we forecast emissions SO2 for a period of one to three months. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2019.4.03 |
| first_indexed | 2025-07-17T10:26:26Z |
| format | Article |
| fulltext |
S.I. Zabielin, 2019
30 ISSN 1681–6048 System Research & Information Technologies, 2019, № 4
UDC 004.8
DOI: 10.20535/SRIT.2308-8893.2019.4.03
FORECASTING SO2 EMISSION OF KILAUEA VOLCANO USING
INTELLIGENT METHOD OF DATA ANALYSIS
S.I. ZABIELIN
Abstract. Kilauea is one of the most active and well-known volcanoes in the world
and most of our knowledge of volcanism originates from its research. During a long
study of volcanoes, many different methods of forecasting their activity were pro-
posed, from the seismological analysis to the statistical analysis of their emissions.
However, a comprehensive analysis of data arrays with the help of intelligent meth-
ods of data analysis has not been carried out before. Using fuzzy data processing
methods, a neural network, volcanic and atmospheric indicators, we forecast emis-
sions SO2 for a period of one to three months.
Keywords: neural network, volcanology, fuzzy logic, LSTM.
INTRODUCTION
Kilauea is a shield volcano on the island of Hawaii on the southeastern tip of the
Hawaiian Archipelago. It is one of the most active and well-known volcanoes in
the world and most of our knowledge of volcanism originates from his research.
Kilauea often erupted and expanded over a long period of time.
Kilauea is one of the most studied volcanoes in the world. Since the arrival
in its neighborhood of the first Christian missionaries in 1823, detailed descrip-
tions of the eruption of the volcano were conducted. In 1912, after the construc-
tion of the Hawaiian Observatory of Volcanoes on the caldera Kilauea, ongoing
scientific research was carried out [1]. Today, this observatory has become one of
the world's leading centers of volcanological research.
During a long study of volcanoes, many different methods of forecasting
their activity were proposed, from seismological analysis to statistical analysis of
their emissions. However, a comprehensive analysis of data arrays with the help
of intelligent methods of data analysis has not been carried out before.
MODEL FOR VOLCANIC ERUPTIONS
Many different methods of volcanology try to predict the behavior of volcanoes
using temperature data or seismic models [2], lava motion models, or gas motion
models. The proposed model used to predict volcanic activity consists of three
components: volcanological, atmospheric, and seismic models (Fig. 1).
During volcanic eruptions, a hot mixture of particles and volcanic gases is
typically released into the air. The main parameters of the volcanological model
include indicators of carbon dioxide (tons per month), sulfur dioxide (tons per
month) and a discrete volcanic explosivity index from 0 to 8, where 0 to
Forecasting SO2 emission of Kilauea volcano using intelligent method of data analysis
Системні дослідження та інформаційні технології, 2019, № 4 31
10,000 meters of cubic ejected materials, 8 – more than 1,000 cubic kilometers of
ejected materials [3].
Emissions of gases and hot particles affect the atmospheric performance of
the metrological stations closest to the volcano. The main parameters of the at-
mospheric model include temperature anomaly (°C/ month), precipitation (mm
per month), air pressure, wind speed. Thus, an increase in the temperature anom-
aly precedes the eruption of a volcano and can provide the necessary data for
forecasting [4].
The eruption can also be preceded by seismic activity several hours or days
before the eruption. The main parameters of the seismic model is the relative
magnitude of soil vibrations [5].
STATEMENT OF THE PROBLEM
Task is to predict the behavior of the volcano using the volcanic eruption model
described above and various data collected from volcanic studies over a long pe-
riod.
Data
The first component is the data. It was taken from a different number of sources
and reports [6–9]. Data can be divided into several categories.
Volcanological indicators include:
1. Carbon dioxide CO2. The gas that is released during the volcanic activity
of a volcano.
2. Sulfur dioxide SO2. A colorless gas with a pungent odor that irritates the
skin and mucous membranes of the eyes, nose and throat.
Fig. 1. Volcanic Eruption Model
Seismic activity
Atmospheric indicators
Volcanological
emissions
S.I. Zabielin
ISSN 1681–6048 System Research & Information Technologies, 2019, № 4 32
3. Indicator of volcanic activity. An indicator of the strength of a volcanic
eruption based on an estimate of the volume of erupted products and the height of
the ash column.
These indicators directly show the activity of the volcano. With a sharp in-
crease in emissions, it can be concluded about the onset of volcanic activity.
Atmospheric indicators include:
4. Temperature anomaly (°C /month). Deviation from the reference value or
long-term average.
5. Precipitation (mm per month).
6. Air pressure.
7. Wind speed. Allows you to determine the speed of gas propagation from
the bowels of the volcano and make the appropriate correction for volcanological
indicators.
Seismic indicators include.
Relative magnitude of soil vibrations, calculated by the following formula:
2d
M
M r ,
where M is the magnitude of the earthquake at the epicenter and d is the dis-
tance from the epicenter to the volcano.
METHODS USED
To analyze data, fuzzy logic and neural networks were used.
Echo time series
For higher accuracy of the neural network, the echo time series have been devel-
oped. This method is used for time series with high sparseness. Instead of fading
immediately, a surge in the value in the time series generates an echo that propa-
gates further along the time series. Each echo term in a time series is described by
the following formula:
ni
n
iiii xkxkkxxE 2
2
1 ,
where iE is the echo value of the time series, ix is the value of the input time
series at time ki, is the attenuation coefficient )1( k , n is the attenuation limit,
which determines the limit of the influence of previous values on the current
value.
This method was used in the current problem for the relative soil vibration
magnitudes, since earthquakes are rare and sporadic.
Fuzzification
To use fuzzy data it was fuzzified. Fuzzification is the process of changing a real
scalar value into a fuzzy value. This is achieved with the different types of fuzzi-
fiers (membership functions) [10].
Forecasting SO2 emission of Kilauea volcano using intelligent method of data analysis
Системні дослідження та інформаційні технології, 2019, № 4 33
Fuzzy data were presented in the form of triangular fuzzy numbers, de-
scribed using three numbers — },,{ cba , where b is the mode of the fuzzy num-
ber, a and c are the degrees of fuzziness of the fuzzy number. They were calcu-
lated from initial data using the following formulas:
ii xb ,
N
N
xa ii
1
,
N
N
xc ii
1
,
where ix is the initial clear data at time i, N is the number of input variables, for
this problem 8N .
Thus, using the above-described method of fuzzification, we can obtain
fuzzy numbers, which, with an increase in the number of input parameters, tend to
real scalar data. This consideration is justified by the assumption that the forecast
will be less fuzzy using all the indicators of the data. In other words if we had the
opportunity to obtain the location and velocity vector of all molecules within a
radius of 10 km from the volcano, then the forecast could be given with undeni-
able accuracy.
Neural networks
To solve this problem, it was decided to use recurrent neural networks, namely
long-term short-term memory (LSTM). Like most recurrent neural networks, an
LSTM network is universal in the sense that, with a sufficient number of network
elements, it can perform any calculation that a regular computer is capable of,
which requires an appropriate weight matrix that can be considered as a program.
Unlike traditional recurrent neural networks, the LSTM network is well suited for
training on the problems of classification, processing and forecasting time series
in cases where important events are separated by time lags with an indefinite du-
ration and boundaries. The relative immunity to the duration of time gaps gives
LSTM an advantage over alternative recurrent neural networks, hidden Markov
models, and other training methods for sequences in various fields of application [11].
RESULTS AGGREGATION PROCESS
To predict volcanic activity, we used the process of aggregation of results, pre-
sented in Fig. 2.
Fig. 2. Schematic process of aggregating results
S.I. Zabielin
ISSN 1681–6048 System Research & Information Technologies, 2019, № 4 34
The input data that was described earlier goes through several stages of data
conversion, fuzzification, forecasting and de-fuzzification. Each part of the proc-
ess receives input from the previous stage and, after appropriate transformations,
transfers it to the next stage.
Below in Fig. 3, 4 and Table, you can see an example of the aggregation
process with intermediate values obtained at various stages of its operation.
Fig. 3. Example of an aggregation process for SO2 emissions
Fig. 4. Example SO2 Emissions Aggregation Process
Forecasting SO2 emission of Kilauea volcano using intelligent method of data analysis
Системні дослідження та інформаційні технології, 2019, № 4 35
An example of forecasting
Date Input SO2
Fuzzy
Low
Fuzzy
High
Forecasted
Low
Forecasted
High
Scalar
Result
MPE
1983,00 1670,20 835,10 2505,30 1632,10 1765,20 1698,65 0,04
1983,08 1872,33 936,17 2808,50 1193,17 2347,33 1770,25 0,25
1983,17 1944,51 972,26 2916,77 1415,26 2753,51 2084,38 0,24
1983,25 1864,81 932,40 2797,21 1091,40 2090,81 1591,11 0,24
1983,33 1233,60 616,80 1850,41 816,80 1422,60 1119,70 0,21
1983,42 1560,11 780,05 2340,16 1189,05 2202,11 1695,58 0,23
1983,50 1552,24 776,12 2328,35 444,12 1856,24 1150,18 0,38
1983,58 1229,47 614,74 1844,21 966,74 1140,47 1053,60 0,08
1983,67 1740,36 870,18 2610,53 1026,18 2327,36 1676,77 0,28
1983,75 1680,17 840,08 2520,25 1015,08 2161,17 1588,13 0,27
1983,83 2066,06 1033,03 3099,09 1411,03 1964,06 1687,55 0,14
1983,92 2157,86 1078,93 3236,79 1065,93 2583,86 1824,90 0,29
FORECASTING RESULTS
Prediction was carried out for emissions of SO2 for a period of one to three
months.
The predicted indicators are a triangular fuzzy number, where each of the pa-
rameters is calculated by a separate neural network. Each of the three neural net-
works receives real scalar data, the first is the lower boundary of the triangular
number, the second is the mode, and the third is the upper boundary.
The result of the work of these networks and their various training became
predicted fuzzy number. Cross-validation of data was performed when dividing
the data into a training and verification sample. Data was taken from 1980 to
1997. The results are given below.
As can be seen when forecasting at 1 month (fig. 5–6), good forecasting re-
sults were obtained with an original SO2 level that stayed in the boundaries of the
predicted values.
СONCLUSIONS
In this paper, we presented a method for predicting SO2 emissions using a large
number of diverse input data, phasing methods, recurrent neural networks, and
time series echoes. High accuracy indicators were obtained when forecasting from
one to three months in advance.
This method can be used to detect volcanic activity at an early stage of its
initiation and to prevent the catastrophic consequences that volcanic activity can
bring with it.
S.I. Zabielin
ISSN 1681–6048 System Research & Information Technologies, 2019, № 4 36
1
3
2
F
ig
. 5
. F
or
ec
as
ti
ng
e
m
is
si
on
s
fo
r
1
m
on
th
: 1
—
S
O
2
(O
ri
gi
na
l)
;
2
—
S
O
2
(L
ow
);
3
—
S
O
2
(M
ed
);
4
—
S
O
2
(H
ig
h)
S
O
2
1
M
on
th
F
or
ec
as
t
Forecasting SO2 emission of Kilauea volcano using intelligent method of data analysis
Системні дослідження та інформаційні технології, 2019, № 4 37
F
ig
. 6
. F
or
ec
as
ti
ng
e
m
is
si
on
s
fo
r
3
m
on
th
: 1
—
S
O
2
(O
ri
gi
na
l)
;
2
—
S
O
2
(L
ow
);
3
—
S
O
2
(M
ed
);
4
—
S
O
2
(H
ig
h)
3
1
2
4
S
O
2
3
M
on
th
F
or
ec
as
t
S.I. Zabielin
ISSN 1681–6048 System Research & Information Technologies, 2019, № 4 38
СONCLUSIONS
In this paper, we presented a method for predicting SO2 emissions using a large
number of diverse input data, phasing methods, recurrent neural networks, and
time series echoes. High accuracy indicators were obtained when forecasting from
one to three months in advance.
This method can be used to detect volcanic activity at an early stage of its
initiation and to prevent the catastrophic consequences that volcanic activity can
bring with it.
REFERENCES
1. Hawaii Volcanoes National Park (N.P.), natural resources management plan. — Ha-
waii: Department of the Interior, 1974. — P. 32.
2. Shimozuru D. A seismological approach to the prediction of volcanic eruptions /
D. Shimozuru // The Surveillance and Prediction of Volcanic Activity. — Paris,
1971. — P. 19–45.
3. Aiello G. Volcanoes: geological and geophysical setting, theoretical aspects and nu-
merical modeling, applications to industry and their impact on the human health /
G. Aiello. — London: IntechOpen, 2018. — 285 p.
4. Gvishiani A.D. Artificial intelligence and dynamic systems for geophysical applica-
tions / A.D. Gvishiani, J.O. Dubois. — Berlin: Springer, 2011. — P. 239–283.
5. Zobin V.M. Introduction to volcanic seismology / M. V. Zobin. — Amsterdam:
Elsevier, 2017. — P. 29–43.
6. Elias T. Sulfur dioxide emission rates of Kilauea Volcano, Hawaii, 1979–1997.
Menlo Park, CA: U.S. / T. Elias // Geological Survey. — 1998.
7. Carey R. Hawaiian volcanoes: from source to surface / R. Carey. — Washington,
D.C: American Geophysical Union, 2015. — P. 393–404.
8. Poland M.P. Characteristics of Hawaiian volcanoes / M.P. Poland, T.J. Takahashi,
C.M. Landowski. — Reston, Virginia: U.S. Department of the Interior, U.S. Geo-
logical Survey, 2014. — 429 p.
9. Helz R.T. Whole-rock analyses of core samples from the 1988 drilling of Kilauea Iki
lava lake, Hawaii / R.T. Helz, J.E. Taggart // Open-File Report. — 2010. — doi:
10.3133/ofr20101093
10. Bhargava A.K. Fuzzy set theory fuzzy logic and their applications / A.K. Bhargava //
Ram Nagar, New Delhi: S CHAND & CO LTD, 2013. — P. 315–348.
11. Goodfellow I. Deep learning. Cambridge / I. Goodfellow, Y. Bengio, A. Courville.
— MA: MIT Press, 2017. — P. 408–412.
Received 25.10.2019
From the Editorial Board: the article corresponds completely to submitted manuscript.
|
| id | journaliasakpiua-article-181465 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:26:26Z |
| publishDate | 2019 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/d6/ade10efbbcd1cb7a2ee5173bcd20aad6.pdf |
| spelling | journaliasakpiua-article-1814652020-03-02T17:05:10Z Forecasting SO2 emission of Kilauea volcano using intelligent method of data analysis Прогнозирование эмиссии SO2 вулкана Килауэа с использованием интеллектуального метода анализа данных Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних Zabielin, Stanislav neural networks volcanology fuzzy logic LSTM нейронні мережі вулканологія нечітка логіка LSTM нейронные сети вулканология нечеткая логика LSTM Kilauea is one of the most active and well-known volcanoes in the world and most of our knowledge of volcanism originates from its research. During a long study of volcanoes, many different methods of forecasting their activity were proposed, from the seismological analysis to the statistical analysis of their emissions. However, a comprehensive analysis of data arrays with the help of intelligent methods of data analysis has not been carried out before. Using fuzzy data processing methods, a neural network, volcanic and atmospheric indicators, we forecast emissions SO2 for a period of one to three months. Килауэа является одним из самых активных и известных вулканов в мире, и большая часть знаний о вулканизме основывается на исследованиях. Во время длительного изучения вулканов было предложено много различных методов прогнозирования их активности от сейсмологического анализа до статистического анализа их выбросов. Однако комплексный анализ массивов данных с помощью интеллектуальных методов анализа данных ранее не проводился. С использованием нечетких методов обработки данных, нейронной сети, вулканических и атмосферных показателей спрогнозированы выбросы SO2 на период от одного до трех месяцев. Кілауеа є одним з найактивніших і відомих вулканів у світі, і велика частина знань про вулканізм ґрунтується на його дослідженнях. Під час тривалого вивчення вулканів запропоновано багато різних методів прогнозування їх активності, від сейсмологічного аналізу до статистичного аналізу їх викидів. Однак комплексний аналіз масивів даних за допомогою інтелектуальних методів аналізу даних раніше не проводився. Із використанням нечітких методів оброблення даних, нейронної мережі, вулканічних і атмосферних показників спрогнозовано викиди SO2 на період від одного до трьох місяців. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2019-12-23 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/181465 10.20535/SRIT.2308-8893.2019.4.03 System research and information technologies; No. 4 (2019); 30-38 Системные исследования и информационные технологии; № 4 (2019); 30-38 Системні дослідження та інформаційні технології; № 4 (2019); 30-38 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/181465/189973 Copyright (c) 2021 System research and information technologies |
| spellingShingle | нейронні мережі вулканологія нечітка логіка LSTM Zabielin, Stanislav Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних |
| title | Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних |
| title_alt | Forecasting SO2 emission of Kilauea volcano using intelligent method of data analysis Прогнозирование эмиссии SO2 вулкана Килауэа с использованием интеллектуального метода анализа данных |
| title_full | Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних |
| title_fullStr | Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних |
| title_full_unstemmed | Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних |
| title_short | Прогнозування емісії SO2 вулкана Кілауеа з використанням інтелектуального методу аналізу даних |
| title_sort | прогнозування емісії so2 вулкана кілауеа з використанням інтелектуального методу аналізу даних |
| topic | нейронні мережі вулканологія нечітка логіка LSTM |
| topic_facet | neural networks volcanology fuzzy logic LSTM нейронні мережі вулканологія нечітка логіка LSTM нейронные сети вулканология нечеткая логика LSTM |
| url | https://journal.iasa.kpi.ua/article/view/181465 |
| work_keys_str_mv | AT zabielinstanislav forecastingso2emissionofkilaueavolcanousingintelligentmethodofdataanalysis AT zabielinstanislav prognozirovanieémissiiso2vulkanakilauéasispolʹzovaniemintellektualʹnogometodaanalizadannyh AT zabielinstanislav prognozuvannâemísííso2vulkanakílaueazvikoristannâmíntelektualʹnogometoduanalízudanih |