improved algorithms for using radon emission

IMPROVED ALGORITHMS FOR USING
RADON EMISSION ANOMALIES AS
RELIABLE PRECURSORS TO EARTHQUAKES,
USING STATISTICAL AND ARTIFICIAL
NEURAL NETWORKS
DHAWAL GUPTA
INSTRUMENT DESIGN DEVELOPMENT CENTER
INDIAN INSTITUTE OF TECHNOLOGY DELHI
JUNE 2012
© Indian Institute of Technology Delhi (IITD), New Delhi, 2012
IMPROVED ALGORITHMS FOR USING
RADON EMISSION ANOMALIES AS
RELIABLE PRECURSORS TO EARTHQUAKES,
USING STATISTICAL AND ARTIFICIAL
NEURAL NETWORKS
by
DHAWAL GUPTA
INSTRUMENT DESIGN DEVELOPMENT CENTER
Submitted
in fulfillment of the requirement of the degree of
Doctor of Philosophy
to the
Indian Institute Of Technology Delhi
June 2012
Dedicated to Almighty
for providing me great teachers and
loving family
Certificate
This is to certify that the thesis entitled “Improved Algorithms for using Radon
Emission Anomalies as Reliable Precursors to Earthquakes, using Statistical and
Artificial Neural Networks” being submitted by Dhawal Gupta for the award of the
degree of the Doctor of Philosophy to the Indian Institute of Technology, Delhi, is a
record of the bonafide research work he has carried out under our supervision. The
results contained in this thesis have not been submitted to any other University or
Institute for the award of a degree or diploma.
(Dr. D.T. Shahani)
Professor
Instrument Design Development Center
Indian Institute of Technology, Delhi
New Delhi - 110016
India
(Dr. Shakeb A Khan)
Associate Professor
Electrical Engg. Deptt.
Jamia Millia Islamia
New Delhi - 110025
India
Acknowledgements
I would like to convey my sincere gratitude to Prof. D.T. Shahani for his
invaluable guidance, constant encouragement and support throughout this work.
No amount of words can really be adequate in expressing my sincere thanks to
him. Without his unremitting support, this work would have never been
completed.
I am greatly indebted to Dr. B.K. Bansal, Advisor, Ministry Of Earth
Sciences, for providing me inspiration and support throughout the course of
research.
I am especially grateful to Dr. N.K. Jain, Head, IDDC, and to the other
faculties, students and staff members of the IDDC, for standing by my side
through my tiring times and providing immense encouragement and unsolicited
support. I am also thankful to Ms. Jasleen Kaur and Mr. M.S. Negi, MDIT Lab,
IDDC for their help.
My acknowledgement would be incomplete if I forget to thank my wife,
Anchal Gupta for her love and support that has seen me through.
I would finally like to thank all those directly and indirectly involved in the
making of this thesis and my research work.
DHAWAL GUPTA
Abstract
Earthquake prediction is one of the problems which is always on the horizon of
seismology. Being a physical phenomenon, many techniques used currently for prediction
purposes are based on geophysical approaches, like seismology, magnetism, electricity, and
geodesy. For India, Seismic hazard evaluation in tectonically active Himalayas, the meeting
region of Indian & Eurasian plates is crucial because earthquakes pose a continual threat to the
safety of the people inhabiting this young and gigantic mountain system of the world. It is
essential that scientific research based on earthquake precursors be carried out in order to
generate earthquake warning sufficiently ahead of the event, be significantly accurate and have
low false alarms.
However, following the reviews on earthquake prediction, it can be observed that the
present day methods for detecting precursory phenomena that precede large earthquakes
additional work is needed to improve true predictions and reduce false alarms, and also be
applied in more systematic and reliable way. Indeed, for numerous precursory phenomena
those have been identified subsequent to many earthquakes, the statistically based reliable data
for the recognition of a method based on the search for precursor’s calls for lot of
improvement.
The investigation throughout the world in past two decades provides evidence which
indicates that significance variation of radon and other soil gases may occur in association with
major geophysical events such as earthquake events. Therefore, the investigation of variation
of soil gases, especially radon being used as a earthquake precursor in present study, can
certainly generate valuable information that can be quite useful in the predictive earthquake
modeling study of tectonic activities within the earth’s crust.
i
Emission of Radon is strongly influenced by day to day meteorological conditions as
well as seasonal. To detect earthquake related anomaly, the raw data on radon emission needs
to be corrected for these interfering influences. The day to day meteorological influences on
emitted radon have been tackled by some form of regression / corrections on raw data of
emission based on measured meteorological parameters like humidity, temperature, pressure
etc.. An improvement in removing variations in emission due to these is desirable and has been
presented in this thesis.
The thesis has two main parts. In the first part of the thesis, the traditional statistical
algorithm which included regression to remove the effect of the meteorological parameters
mainly barometric pressure and rainfall from the as is measured radon has been presented with
an additional variation that periodicity in seasonal variations as computed using FFT is shown
to improve reliability of prediction of earthquake and it is shown that the method presented in
the thesis leads to a more generalized approach for use at different locations. The second part
deals with the use of neural network algorithms which can learn the behavior of radon with
respect to known meteorological parameters. This method has potential of tracking “changing
patterns” in dependence of radon on meteorological parameters and it may adapt to such
changes on its own in due course of time. Here also use of periodicity obtained by FFT is
shown to give better results. Another method that requires neither an explicit step of regression
nor use of any specific period is also presented. The method involves use of Probabilistic
Neural Networks that take all possible measured data (like emitted radon, meteorological
conditions) as inputs and focus on earthquake events as final output. All the methods
developed in the thesis have used radon data in the tectonically active Himalayas from the
established radon measurement stations in India.
The relative efficacy of the newly developed algorithms has been compared in terms of
TA (True Anomalies) and FA (False Anomalies) for use of Radon Emission Anomaly as
ii
precursor for predicting earthquake. It is shown that use of specific periods of analysis as given
in the thesis, and correction methods on raw data of emitted radon, result in significant increase
in TA prediction and reduction in FA. The “learning” capability of neural networks has also
been exploited for improved reliability in earthquake prediction. The algorithms presented in
the thesis are easier to implement in newer locations and can find applications in other types of
earthquake precursors.
iii
CONTENTS
List of Figures…………………………………………………………………………….vi
List of Tables……………………………………………………………………………..vii
Chapter 1: INTRODUCTION AND REVIEW OF LITERATURE
1.0
Introduction
1
1.1
Earthquake Precursory Studies
2
1.2
Origin, behavior and migration of the soil gases
8
1.3
Soil gas surveys: literature review
9
1.4
Objectives of the thesis
12
1.5
Organization of the thesis
17
Chapter 2 : AN OVERVIEW OF STATISTICAL AND ARTIFICIAL NEURAL
NETWORK TECHNIQUES
2.0
Introduction
19
2.1
Introduction to Statistical Techniques
19
2.1.1
Time Series Analysis
20
2.1.1.1 Smoothing Time Series and Curve (Function) Fitting
21
2.1.1.2 Accuracy of Forecast
22
2.1.2
Nonlinear Estimation
23
2.1.3
Fast Fourier Transforms (FFT)
25
2.1.4
Anomaly Detection:
25
2.1.4.1 Anomaly detection techniques
26
2.1.4.2 Evaluation of anomaly detection technique
28
2.2
Introduction to Neural Networks
28
2.2.1 Basic Architecture and Activation functions
31
2.2.2
Neural Network Topologies
32
2.2.2.1
Single-Layer Feed forward Networks
32
2.2.2.2
Multilayer Networks
33
iv
2.2.2.3
Radial basis Function Networks
35
2.2.2.4
Randomly Connected Networks
37
2.2.2.5
Probabilistic Neural Networks
38
2.3 Artificial Neural Network Learning
39
2.4 Summary
40
Chapter 3: STATISTICAL ANALYSIS OF RADON AS AN EARTHQUAKE
PRECURSOR
3.0 Introduction
42
3.1 Radon Measurement Technique
45
3.2 Algorithm for using radon emission anomaly as earthquake Precursor
47
3.3 Regression and Correction (Normalization) of the Measured Radon
48
3.4 Results and Comparison of Proposed Algorithm
58
3.5 Summary
63
Chapter 4: NEURAL NETWORK ANALYSIS OF RADON AS AN EARTHQUAKE
PRECURSOR
4.0 Introduction
65
4.1 Regression and Correction of the Measured Radon approach
65
4.2 Neural Networks in Radon Emission Study
66
4.3
4.2.1
Neural Network algorithm for radon emanations estimate
68
4.2.2
Result: Neural Network algorithm for radon emanations estimate
72
4.2.3
Neural Network algorithm for probabilistic event estimate
77
Summary
81
Chapter 5: CONCLUSION AND FUTURE WORK
5.0 Earthquake precursor modeling
83
5.1 Main Conclusions
87
5.2 Possibilities for Future Work
88
References
90
Bio-Data
97
v