Transcript tissot-longbeach-mar - Division of Nearshore Research

Texas A&M University-Corpus Christi  Division of Near Shore Research
Present Trends in the Application
of Artificial Intelligence to
Environmental Systems
Account of 3rd Conference on AI
Applications to Environmental Science
AI Techniques/Disciplines
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Genetic Algorithms
• Neural Networks
• Support Vector Machines (generalized
version of NNs for image analysis)
• Expert Systems
• Fuzzy Logic
Genetic Algorithms
Texas A&M University-Corpus Christi  Division of Near Shore Research
• GA’s are being used for a number of
environmental applications
–
–
–
–
–
–
Calibrating water quality models
Managing groundwater
Astrophysics
Source of air pollutant
Predatory-Prey model
Inverse problems in general
Genetic Algorithms (2)
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Key to application to Environmental
Science problems is to set the GA as an
optimization problem
• Non linear differential equations can be
approached using GA
Neural Networks
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Increasingly used by NCEP/NOAA/NASA and
European Agencies to speed up computationally
expensive tasks (NN can be orders of
magnitude faster than classic algorithms)
• Use of NNs to correct model biases
• Use of NNs for predictive learning (system
imitation)
Neural Networks (2)
Texas A&M University-Corpus Christi  Division of Near Shore Research
• NN for predictive learning: choosing the “best”
model for a given loss function
• “Do not solve a given problem by indirectly
solving a more general and more complicated
problem as an intermediary step”
• Possible problems for NNs when the size of the
problems become to large (atmospheric models)
General NN Techniques
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Our techniques are very much in line with
other modelers (choice of model parameters,
normalization, model evaluation, etc.)
• Use of PCA analysis to identify smaller more
optimum input decks
• Use of NNs for non linear PCA analysis
Discussion Item
(Caren Marzban)
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Recognition that the parameters of AI models
generally do not have a physical significance
• In particular there is no physical meaning in the
biases and weights of a neural network
• Question/Hypothesis: The parameters of any non
linear models do not provide insight into the
physics of the system
Support Vector Machines
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Relatively new technique which vectorizes
pictures/maps and uses a NN type methodology to
extract features
• uses kernel functions
• Also kernel PCA, kernel Gram-Schmidt, Bayes Point
Machines, Relevance and Leverage Vector Machines,
are some of the other algorithms that make crucial use
of kernels for problems of classification, regression,
density estimation, novelty detection and clustering
Support Vector Machines (2)
Texas A&M University-Corpus Christi  Division of Near Shore Research
• For classification, SVMs operate by finding
a hypersurface which attempts to split the
positive examples from the negative
examples. The split will be chosen to have
the largest distance from the hypersurface to
the nearest of the positive and negative
examples.
Texas A&M University-Corpus Christi  Division of Near Shore Research
http://www.computer.org/intelligent/ex1998/pdf/x4018.pdf
Texas A&M University-Corpus Christi  Division of Near Shore Research
http://www.computer.org/intelligent/ex1998/pdf/x4018.pdf
General Issues
Texas A&M University-Corpus Christi  Division of Near Shore Research
• AI techniques are increasingly used to make short
to medium term forecasts (sometimes fractions of
hours) when large amounts of data are available
• Usual forecasting methodologies have trouble for
short term forecasts
• Harris Corporation is implementing a ceiling and
visibility fuzzy logic tool for the FAA
Groundwater Issues
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Population growth and possible global change
affects/will affect potable water supply
including groundwater supply
• Decision making is at the (very) local level and
relatively uncoordinated (likelihood of
conflicts)
• It takes 20-25 years to go from the planning to
the realization of a new reservoir
Bush Administration on
Climate Change
Texas A&M University-Corpus Christi  Division of Near Shore Research
• Global climate change recognized as a capstone
issue of our generation by the Bush administration
• Climate change program ~ 1.7 bio/year
• 4 part focus:
–
–
–
–
Science
Observation and Data
Decision support resources
Outreach and Education
• Core solutions to the problem will be
technological breakthroughs
Scatterometry Overview:
Ocean Circulation
Scatterometer high resolution, extensive, and
frequent wind velocity measurements are
Texas A&M University-Corpus Christi  Division of Near Shore Research
used to understand
upper ocean circulation
from regional to global scales
• Wind stress is the largest momentum
input to ocean
• Wind stress curl drives large-scale
upper ocean currents
• Small-scale wind variability modifies
large-scale ocean circulation
• Coastal regions exhibit amplified
physical/biological response
• Wind forcing complements dynamic
and thermodynamic response
measurements
Winds & Water, Weather & Waves: Satellite Scatterometry, M.H. Freilich, AMS Short Course, February, 2003
Scatterometry Overview:
Meteorology
All-weather surface wind measurements provide
Texas A&M for
University-Corpus
Christi  Divisionresearch
of Near Shore Research
unique information
atmospheric
and operations
Winds & Water, Weather & Waves: Satellite Scatterometry, M.H. Freilich, AMS Short Course, February, 2003
Hurricane Dora
QuikSCAT -- 10 August 1999
Global backscatter and ocean vector winds
< 11 hour coverage
Texas A&M University-Corpus Christi  Division of Near Shore Research
M. H. Freilich COAS/OSU
W. L. Jones
UCF
Winds & Water, Weather & Waves: Satellite Scatterometry, M.H. Freilich, AMS Short Course, February, 2003
WIND MEASUREMENTS:
Mission Schedules
Texas A&M University-Corpus Christi  Division of Near Shore Research
99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15
[SSM/I, SSMIS]
QSCAT
SWS/ADEOS-II
SWS/GCOM-B
ERS-2
ASCAT/METOP
WINDSAT
CMIS/NPOESS