an evidence science approach to volcano hazard forecasting
TRANSCRIPT
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
1/28
EXPLORISMontserratVolcano ObservatoryAspinall and Associates
Risk Management Solutions
1
2
3
4
5
An Evidence Science approach to
volcano hazard forecasting
Thea Hincks1, Willy Aspinall1,2, Gordon Woo3, Gillian Norton4,5
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
2/28
Evidence science
Evidence-based medicine is the conscientious, explicit and judicioususe of current best evidence in making decisions
the integration of individual expertise with the best available external
evidence from systematic researchAfter Sackett et al., 1996Evidence Based Medicine
Need to model uncertainty and make
forecasts using
Expert judgment & knowledge of
physical system
Observational evidence
= highly complex system
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
3/28
Bayesian networks
Bayesian belief networks (BBNs)Causal probabilistic network
Directed acyclic graph
Set of variables Xi
discrete or continuous
Set of directed links
Variables can represent hidden or
observable states of a system
Very useful in volcanology-our observations on internal
dynamics of the volcano are
indirect
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
4/28
Expert systems
NASA data analysis
MSOffice assistant
Bayesian
Network
applications
Speech recognition
Molecular Biology
and Bioinformatics
Medical diagnosis& decision making
VOLCANIC
HAZARD
FORECASTING
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
5/28
Building a Bayesian network
Sensor model: Prior and transition models
Probability of observation
P(Y|X)
Probability of initial state P(X0)
Transition between states P(X1|X0)
Bayes theorem
P(A |B,C) P(B |A,C)P(A |C)
P(B
|C
)
Filtering - estimate current state XtPrediction - future states Xt+n
Forward pass :
Smoothing-past unobserved states
Backward pass :
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
6/28
Network structure Judgment, physical models, observations
factors we believe lead to instability
Structure learning algorithms
purely data driven model
difficult to model unobserved nodes
problem is NP-hard
algorithms slow to compute
(~ few days for 6 x ternary node graph)
BN for dome collapse on Montserrat
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
7/28
rainfall ondome dome collapse
magma flux
ground
deformation
stability
of edifice
degassing
pressure
Factors that might lead to dome collapse:
BN for dome collapse on Montserrat
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
8/28
rainfall ondome dome collapse
magma flux
ground
deformation
degassing
stability
of edifice Cant measurestate directly hidden variablespressure
BN for dome collapse on Montserrat
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
9/28
magma flux
deformation
SO2 flux
observed rainfallUEA & MVO rain gauges
degassing
stability
pressure
GPS, EDM and tilt
Seismicity: VTearthquakes
Long periodearthquakes
Hybrid
Rockfall
LP Rockfall
BN for dome collapse on Montserrat
use sensor models forour observations:
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
10/28
Data
Testing with daily data from July 95 - August 04
S02 flux
Ground deformation (4 GPS lines) 4 nodes
Seismic activity (event triggered count & magnitudedata) VT, Hybrid, LP, LPRF, RF 5 nodes
Rainfall
Collapse activity
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
11/28
Time dependence
Structure: how are processes coupled?
What is the order of the process ?
Dynamic system- history is important Variables tied over several time slices
Time ser ies analys is of m oni tor ing data
Autocorrelation & partial autocorrelation functions, differenced data
Approximate order for time dependent processes
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
12/28
Autocorrelations
Computed
autocorrelationfunction and and
partial
autocorrelation
function for data
and firstdifferenced data
check structureis sensible and
est imate order of
t ime dependence
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
13/28
Dynamic Bayesian Network
Rainfall - 1 day autocorrelation
Hidden Markov model O(1)
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
14/28
Dynamic Bayesian Network
Pressure
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
15/28
Dynamic Bayesian Network
Magma flux
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
16/28
Dynamic Bayesian Network
Gas flux
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
17/28
Dynamic Bayesian Network
Ground deformation
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
18/28
Dynamic Bayesian Network
Structural integrity or stability
of the dome is dependant on
previous state
prior rock fall activity
prior collapse activity(also affects pressurization)
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
19/28
Dynamic Bayesian Network
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
20/28
Current model
Where monitoring time series suggest higher order processes
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
21/28
Current model
Prior distribution
Expert judgment
Sensor model
Transition model
Expert judgment to set initial
distributions Parameter learning algorithms
on monitoring data
P(X0), P(Y0)
for all states X
observations Y
P(Yt|Xt)
P(Xt+1|Xt)
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
22/28
Results so far
Parameter learning using ~9 years of data
transition and sensor models
1. static BN2. two-slice dynamic model
3. three-slice dynamic model
Can estimate probability of collapse given new observations
Smoothing to estimate hidden state probabilities and distributions formissing values of observed nodes
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
23/28
Results so far
Structure learning on a small (5 node) model - observed nodes only
work still in progress!
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
24/28
Results so far
High ground deformation
Consistent, moderate hybrid activity
No SO2 observations
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
25/28
Results so far
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
26/28
Further work
Model observations with continuous nodes More monitoring data - extend network
Look at full seismic record (not just event triggered
data)
Run structure learning algorithm on larger network
Investigate second order uncertainties (model
uncertainty) and scoring rules to see how well
different models perform
User interface for real time updating of network at
MVO real time forecasting probability of collapse
Longer range forecasting?
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
27/28
Conclusions
All models are wrong (to some degree)but some models are better than others
EVIDENCE SCIENCE and BAYESIAN NETWORKSRobust, defensible procedure for combining
observations, physical models and expert judgment
Risk informed decision making
Can incorporate new observations/phenomena as they occur
Strictly proper scoring rules - unbiased assessment of
performance & model uncertainty
-
7/27/2019 An Evidence Science approach to volcano hazard forecasting
28/28
References
Druzdzel, M and van der Gaag, L., 2000. Building Probabilistic Networks:Where do the numbers come from? IEEE Transactions on Knowledge
and Data Engineering 12(4):481:486
Jensen, F., 1996.An Introduction to Bayesian Networks. UCL Press.
Matthews, A.J.and Barclay J., 2004A thermodynamical model for rainfall-
triggered volcanic dome collapse. GRL 31(5)
Murphy, K., 2002Dynamic Bayesian Networks: Representation, Inference
and Learning. PhD Thesis, UC Berkeley. www.ai.mit.edu
openPNL(Intel) http://sourceforge.net/projects/openpnl
open source C++ library for probabilistic networks/directed graphs