Complex stochastic approach for prediction of natural catastrophic events: earthquakes and volcanic

1
Complex stochastic approach for prediction of
natural catastrophic events earthquakes and
volcanic eruptions.

• ALEXANDER ZORIN
• Saint Petersburg State University,
• Faculty of Mathematics and Mechanics
• Alexander.Zorin_at_gmail.com

2
Outline

• Problem
• Objectives
• Existing methods of prediction
• Proposed method
• Implementations
• Results
• Accompanying problems and solutions
• Conclusion
• Perspectives future research
• Questions

3
Problem

• How to forecast threat of volcano eruption or
  earthquake in a certain place?
• How to improve reliability of prediction?
• Can we use combined data sources for it?
• How physical model influences on statistical
  quality of forecasting (covariance, robustness of
  estimation)?
• Is there opportunity to apply observational data
  of one place to predict event in another?
• Which properties of prognostic system are
  important for minimization of alarm delay?

4
Objective

• Is to propose a method which will improve
  reliability of a catastrophic event prediction
• Research effective means of combining data
  sources of heterogeneous origin (data fusion)
• Produce technique which suppose to adjust
  physical model settings to observational system
  at its best
• On practice explore appliance of stochastic
  estimation approach for a dynamic analysis of
  geophysical systems
• Develop a method which provide enough time
  reserve for warning/reaction before disaster

5
Existing methods of earthquakes prediction

• Long-term techniques
• Paleoseismology, data about areas recurrence
  intervals
• Seismic gaps, knowledge of potentially
  dangerous zones
• Short-term techniques
• Foreshocks, tremor signals
• Measurable surface displacement, shows strain in
  rocks
• Rapid changes of radon gas concentration in
  natural sources
• Changes in the water-levels of wells
• Changes in the electrical resistivity of rocks
• Unusual radio-waves
• Strange animal behavior

6
Existing methods of volcanic eruptions prediction

• Long-term techniques
• Volcanic history, surface examination,
  radiometric age analysis
• Learning of volcano activity and its eruption
  type suppose to determine its possible behavior
  and dangerous area, and figure hazards map
• Short-term techniques
• Measurements of surface displacement
• Precursor earthquakes, volcanic tremor signals
• S-wave shadow (movement) research
• Measuring changes in magnetic field
  (temperature-dependent)
• Changes in the electrical resistivity of rocks
• Changes in the water-tables (level, temperature,
  admixtures)
• Infrared remote sensing of changes in heat flow
• Changes in gas compositions

7
Existing tools for seismic analysis

• NASA, Earth Observing System (EOS) Science
  Program. GPS sensing, data processing, data
  fusion tools
• Research Center for prediction of earthquakes and
  volcanoes eruption, Tohoku University, Japan.
  Developed observational network, database, data
  processing software (unfortunately documentation
  in Japan only)
• GFZ Potsdam
• GIANT - written by Andreas Rietbrock () - is an
  analysis system for consistent analysis of large,
  heterogeneous seismological data sets. It
  provides a GUI between a relational database and
  numerous analysis tools (such as HYPO71, FOCMEC,
  PREPROC, SIMUL, PITSA, etc. ). The GIANT system
  is currently supported on SunOS, Solaris and
  Linux and uses the X11 windowing system.
• PITSA - in its current version written by Frank
  Scherbaum, Jim Johnson and Andreas Rietbrock - is
  a program for interactive analyis of
  seismological data. It contains numerous tools
  for digital signal processing and routine
  analysis.
• GEOFON Network (Global network) by Dr. Rainer
  Kind
• The Geological Survey of Canada (GSC)'s
  Geodynamics Program, Pacific Geoscience Centre
  (PGC), Sidney, British Columbia, Canada
• Computing tools of Los Alamos Geodynamics
  Laboratory

8
Proposed method. Conceptual scheme
9
Proposed method. Stages

• Environmental model description. State space
  approach
• Observation and estimation error minimization
• Sensor fusion
• Data fusion

10
Proposed method. State space approach (1.1)

• Complex stochastic approach of natural hazards
  prediction in the data assimilation model

11
Proposed method. State space approach (1.2)

• Deterministic state space model for dynamic
  observable variable

12
Proposed method. Observation and estimation error
minimization (2.1)

• Dynamic observable process is regression function

• Deterministic parameter estimation problem.
  Criterion of error between the measurements and
  the model

13
Proposed method. Observation and estimation error
minimization (2.2)

• The same criterion includes a priori information
  about parameter p values

14
Proposed method. Sensor fusion (3.1)

• Dynamics model of the Extended Kalman Filter
  (EKF)

• The measurement model of the Extended Kalman
  Filter (EKF)

15
Proposed method. Sensor fusion (3.2)

• Example of decentralized sensor fusion

16
Proposed method. Data fusion (4.1)

• Information processing of monitoring system and
  decision network (refer to scheme on slide 8)

17
Proposed method. Data fusion (4.2)

• Suitability of observation methods upon
  characteristics for sensor fusion and numerical
  analysis

18
Proposed method. Data fusion. Remarks (4.3)

• The long-term types of forecasting
• They can be implemented in a form of data storage
• It is possible to built data bank for almost all
  selected points of observation area
• The data warehouse contains individual
  information about specific characteristics of a
  point (coordinates, history, geophysical
  properties) so-called pattern
• The short-term techniques
• They based on a similar observations upon several
  dynamic precursor characteristics (temperature,
  energy, )
• It is easy to combine measurements of some
  geophysical precursor characteristics
• In the forecasting activity are differentiated
• Monitoring upon one fixed characteristic in
  several near points (by different sensors)
  sensor fusion
• Monitoring upon a set of different
  characteristics in one fixed point event pattern

19
Data fusion. Requirements of data storage
processing.

• Example
• Seismic-active area of size 10000 square
  kilometers
• 250 distributed points of observation
• 4 precursor characteristics which are taken into
  account
• measurements are taken with frequency 10Hz (each
  0.1second) and stored in an uncompressed format
  during 24 hours
• For each measurement datum is used one element of
  double array (8 Bytes)
• Calculate necessary disc space for raw data

20
Implementations

• Model
• Description of geophysical processes,
• Simulation of environmental behavior dynamics.
• Software for mathematical analysis of
  observational data
• GIS software,
• Sensors measurements processing,
• Estimation and prognosis upon observational data,
  
• Remote sensing improvement and wave propagation.

21
Implementations

• Acquisition data and representation

22
Implementations. Experimental results.

• Conditions of experiment
• Short-term prediction of earthquakes with
  magnitude M 4.0
• Covered area of 8600 square kilometers
• 4 precursor characteristics were observed
• surface displacement,
• amplitude of foreshocks,
• flux variation of terrestrial magnetic field,
• and angle variation of terrestrial magnetic
  field.
• 40 in situ sensors (by 10 for each
• characteristic)
• Duration of experiment 28 days
• Measured shocks and
• calculated probabilities of
• earthquakes

23
Simulation results

• Regression estimation with confidence bounds

24
Simulation results

• Improving measurements by filtration, EKF

25
Accompanying issues and solutions

• Disaster prevention and warning
• Model for selected region
• Connection to the tsunami hazard
• Industrial objects computer-aided design
• Statistic techniques for improving safety of
  geological prospecting
• Application of the technique in mapping and
  routing of oilfields and pipelines dynamic
  charts of hazard probability
• Web-page applet
• For online modeling, real-time observation

26
Conclusion

• Proposed technique includes
• Obtaining measurements and data from different
  kinds of sources
• Estimation and confidence bounds delimitation for
  each source of variety
• Filtering, fuse data, fitting information to the
  pattern, evolution of pattern
• Building the map with determined probability of
  hazard in points
• Making decision of event coming, alarm
  notifications

27
Perspectives future research

• Improve collaboration with existing seismic
  models and event patterns for specified places
• Research the common geophysical sources of
  observed processes to advance particular models
• Apply newest models of filtering
• Implement different styles of data fusion
• Realize new features in C and Matlab and
  compare it with above mentioned programs
• Develop models and tools, which could compete
  with similar projects of other research groups
  (see Slide 7)
• Distributed early warning system based upon
  composite information (both measurements and
  patterns delivered simultaneously),
• Grid-combination of purpose-oriented monitoring,
  modeling and prognostic services
• Finite elements simulation (solid, fluid, mixture
  state)

28
Many thanks!!!

• All for attentive listening
• Organizers for a good company spirit establishing

29
Questions?
30
Metrology Sensorics

• ZnO gas sensors (concentration)
• Precision sensors
• Infrared sensors (T,C)
• Remote sensors (T, emission tracking)-satellite
• Laser sensors (shape and dynamics of earth
  surface)
• Micro Paramagnetic Oxygen Sensor
• etc

31
References

• Amos Nur (2000). "Poseidons Horses Plate
  Tectonics and Earthquake Storms in the Late
  Bronze Age Aegean and Eastern Mediterranean".
  Journal of Archaeological Science 27 4363. ISSN
  0305-4403
• McGuire JJ, Boettcher MS, Jordan TH (2005).
  "Foreshock sequences and short-term earthquake
  predictability on East Pacific Rise transform
  faults". Nature 434 (7032) 445-7. PMID 15791246
• http//pasadena.wr.usgs.gov/step/
• http//gsc.nrcan.gc.ca/geodyn/index_e.php
• http//denali.gsfc.nasa.gov/
• http//ees5-www.lanl.gov/EES5/geo_main.html
• http//www.geodynamics.org/
• http//www.aob.geophys.tohoku.ac.jp/
• http//eospso.gsfc.nasa.gov/

32
PhD preparation

• Time management
• 1y-stage of hypothesis assumption, experiments
  and modeling activity development of simulation
  tools and numerical software.
• 2y-proven results formulation in theses,
  articles, participation in various conferences on
  this stage suppose to have different scientific
  viewpoints on proposed solutions, improving
  experiments
• 3y- publishing papers on new approaches and
  implementing it in industry/environment suppose
  to fix invented methods and prepare well for
  thesis presentation
• Publications
• Scientific contacts and collaborations