John Braun Department
of Statistical and Actuarial Sciences University of Western
Ontario braun@stats.uwo.ca Title: Stochastically Modelling Forest Fire
Spread Forest fires often spread in unpredictable
ways. Deterministic models of fire spread have been developed in both
Canada and the United States. These models capture `expected' behaviour of
fire spread fairly well, but they provide no measures of uncertainty. We
address this problem in two distinct ways. The first way is to propose an
entirely new model which is inherently stochastic. This model is an
interacting particle system on a regular two dimensional lattice. Each
lattice site may be occupied by unburnt fuel, burning fuel, or burnt fuel.
The model evolves according to a continuous time Markov process. An
alternative approach is to enhance the differential equation-based model
that already exists. By applying smoothing and bootstrapping, we can
generate stochastic replicates of fire boundaries. Issues of model
assessment and model acceptance will be discussed. |
David
Brillinger Department of Statistics University of California,
Berkeley brill@stat.berkeley.edu Title (lecture): Probabilistic risk modeling at
the wildland-urban interface: the 2003 Cedar Fire,
III
The October 2003 Cedar Fire in San Diego County
was a tragedy involving 14 deaths, the burning of some 280,000 acres, the
destruction of 2232 homes, and costs of suppression near $30 million, but
the data associated with it provide an opportunity to carry out
probabilistic risk modeling of a wildland-urban interface (WIF). WIFs
exist where humans and their development interface with wildland fuel. As
home building expands from urban areas to nearby rural ones the interface
becomes a greater and greater fire problem.
Since wildfires are an
exceedingly complex phenomenon with uncertainty and unpredictability
abounding a statistical approach to gaining insight appears useful. In
this work spatial stochastic models are developed for relating risk
probabilities and damage measure to various explanatory variables.
There will be discussion of the difficulties that arose in seeking
pertinent data and of carrying out EDA when the data are GIS layers.
The work is collaborative with Benjamin Scott Autrey and Matias
Cattaneo.
Title
(panel discussion): Experience
(no abstract)
|
Mark
Buehner Meteorological Research Division Environment
Canada mark.buehner@ec.gc.ca Title: Data Assimilation for Numerical Weather
Prediction: Estimation and
Modelling of the Covariances of Short-Term
Forecast Error
This presentation describes statistical
estimation techniques used for numerical weather prediction. These
techniques allow large volumes of meteorological data to be combined in an
"optimal" way with short-term forecasts of the atmospheric state from a
numerical model. The resulting estimate of the current atmospheric state
serves as initial conditions for the subsequent model forecast. One major
challenge in applying such techniques is obtaining the probability
distribution of the errors in short-term model forecasts. Typically, only
the covariances of the error are required for most data assimilation
techniques, but even this requires estimating a matrix with on the order
of 10-million-squared elements. Approximations and computationally
efficient approaches for estimating and modelling such covariances will be
discussed, including the use of spatial/spectral localization and
wavelet-like basis functions.
|
Grace Chiu Department
Statistics and Actuarial Science University of
Waterloo gchiu@uwaterloo.ca Title: Where's the Statistician? A Collage of
High Profile Environmental Issues
Public concerns over
the well-being of our environment have been sparked mostly by the few high
profile issues that receive much media attention. Global warming,
pollution, and species extinction for instance, are among the top issues
featured on television and radio, in print, and online. Associated with
them is a myriad of lower profile, but equally pressing, issues arising
due to the interconnectedness of systems and events in nature. What can we
do in the race against time to slow down or reverse the damages that our
species has inflicted on an environment shared by all other species on the
planet? Much of the general public and scientific community alike,
focusses on the advancement in technology and innovative solutions based
on those scientific disciplines deemed immediately relevant to solving
environmental problems, e.g. biology, chemistry, physics, and engineering.
However, little awareness has been given to the vital role that statistics
also should play. Indeed, "you can't fix what you don't understand." Worse
yet, you can't fix something with what you don't understand. And a
thorough understanding of environmental conditions or their alleged
"fixes" would be virtually impossible without involving
statistics.
This presentation will feature some urgent
environmental problems portrayed in the media. The audience is asked to
ponder (1) how often are statisticians involved in these problems, and (2)
how statistics can lead to a better understanding of the problems and
their remedies.
|
Michael
Dowd Department of Mathematics and Statistics Dalhousie
University mdowd@mathstat.dal.ca Title: The Integration of Statistics into
Environmental Prediction Research
This talk will focus
on issues in integrating statistics into marine environmental research.
After some general discussion on the scope of this problem, I will turn to
a specific issue in order to illustrate certain aspects of this problem.
In particular, I examine the case where we have available mathematical
models (often DE based) of the environmental systems of interest. We also
now have the ability to measure (often in near real time) many of the core
environmental variables on the time and space scales of interest, using a
wide variety of measurement technologies. The primary goals for these
models and data are two-fold: (i) to develop a better understanding of how
the environmental system operates; and (ii) to carry out environmental
prediction. The challenge is to develop a class of statistical methods
that can effectively incorporate information from both data and
mechanistic models in order to attain these goals. The first goal is
primarily a scientific one: to test hypotheses, understand and identify
important processes, to investigate system responses to forcing, to guide
sampling strategies, and to identify new research directions. The second
goal is more methodological, and ultimately a validation of our level of
understanding of the system. This talk will discuss the role that
statistics can have in studying these types of environmental systems,
including potential problems and pitfalls. By way of illustration, the
talk will draw heavily from examples taken from my current research
activities. |
Abdel
El-Shaarawi National Water Research Institute Environment
Canada abdel.el-shaarawi@ec.gc.ca Title: Environmental Problems: Identification,
Assessment and Management
When dealing with
environmental problems, understanding the scientific and socioeconomic
dimensions are essential in making the proper identification, assessment
and decisions. Statistical methods are central at all various stages and
this is particularly due to its ability to integrate the diverse knowledge
which is needed to deal with the complex nature of environmental issues.
Routinely, statistics is used by non statisticians to summarize and draw
conclusions from their data. This of course must be encouraged especially
when the methods are properly applied. As history tells us that the roots
of many valuable statistical techniques can trace their origin to the work
of non-statisticians who were dealing with particular applications from
their own field. On the other hand, there are many examples in which
statistical methods are improperly or inefficiently used by
non-statisticians. Similarly statisticians are putting immense efforts in
developing statistical methods that deal with the wrong applied problems
because of not properly understanding the scientific nature of these
problems. The purpose of this presentation is to describe some of the
statistical techniques used in the analysis of water quality data. The
emphases will be on the short comings of these techniques and on how the
collaborations between non-statisticians and statisticians can effectively
enhance each other's contributions in the solution of environmental
problems.
|
Jonathan
Grant Department of Oceanography Dalhousie
University jon.grant@dal.ca Title: Applied Statistics in Ecology and
Biological Oceanography
Ecologists, earth
scientists, and biological oceanographers often have some formal
background in statistics via graduate training. For ecologists, a focus on
experimental manipulations has led to an emphasis on 2-sample or ANOVA as
a staple of community ecology. A leap to other techniques is often limited
as illustrated by several examples. Environmental data is usually
characterized by extreme spatial and temporal variability. Resolution of
this variability is often limited by sampling effort, i.e. manual water or
sediment collection at great ocean depths, through ice, and/or at remote
field sites. However, widespread use of data logging sensors for
fluorescence, temperature, sea level, etc. allows detailed temporal
records to be obtained. There are many pitfalls involving choice of time
series analyses (windowing, normalization, etc.), and there is need for
practical advice from statisticians. In the spatial domain, model results
as well other assessments of spatial data result in various types of maps.
The ability of computers to produce colourful multidimensional maps
is seen as a result in itself, and the map is often believed without
examination of variation in plotted variables. Even in fields with maps as
a core, e.g. GIS and remote sensing, geostatistics is further in the
background that one might expect. Oceanographic models contain
myriad parameters in schemes ranging from simple spreadsheets (index
models) to full blown spatial/temporal simulations. The effect of
variation in coefficients is an often neglected aspect of these
models, despite the availability of tools for risk analysis. Environmental
data and models in the ocean sciences are being used to make important
decisions, even incorporated into formal decision support tools for use by
management. Statistical rigour is often lacking in these systems, despite
an emphasis on supposed risk analysis. Environmetrics has huge
contributions to make to these fields, both in educational foundation for
marine scientists and ecologists, as well as implementation of methods
that improve quantitative performance of data/model presentation,
analysis, and prediction.
|
Timothy G.
Gregoire School of Forestry and Environmental Studies Yale
University timothy.gregoire@yale.edu (discussant: no abstract)
|
Stephen
Murphy Department of Environment and Resource Studies University of
Waterloo sd2murph@fes.uwaterloo.ca Title: Issues in Using Environmetrics in
Ecosystem Research
In "overcoming the challenges," I
will discuss briefly the problems ecologists have with even the
terminology associated with environmetrics. For ecologists, we tend
to focus on advanced statistical analysis and modelling of complex
ecosystem problems. However, a relatively large branch of ecology
uses environmetrics in the sense of statistical and computing techniques
for data mining, warehousing, retrieval and use in broadband and other
forms of data sharing with various communities of practice. My own
research has covered both aspects. I will describe the quantitative
approaches needed for data analysis in the context of multivariate
methods, spatially explicit analyses, and Bayesian approaches. While
some of these are reasonably well established, others are obscure - and
still others are controversial. The problems of coupling these
quantitative analyses with useful approaches to data management is the 2nd
layer of complexity in addressing environmetrics in my areas of ecological
research in urban ecology, ecosystem restoration, and protected
areas/parks management. Consistent with the topics this panel will
address, my approach has been somewhat self-taught with the attendant risk
of having a haphazard education that is incomplete. While I have
worked with statisticians and other ecologists working in the emerging
field of environmetric ecology, we lack systematic education in
undergraduate and graduate training (from current students to professors
like myself) and would benefit from both collaboration and more
professional development workshops.
|
Maren
Oelbermann Department of Environment and Resource Studies University
of Waterloo moelberm@fes.uwaterloo.ca Title: Environmetrics: How a Soil
Scientist Approaches
Statistics
Environmetrics is
the science of understanding ecological and environmental systems through
data, which involves the application and development of statistical
methodologies. On the broad scale, environmental research is
interdisciplinary, and data analyses require an understanding of
statistical design. On a smaller scale, a soil particle, for example
is its own ecosystem with many different components interacting
dynamically. So how does a soil scientist understand these dynamic
interactions in the soil and the interaction between soil and the
environment? Experimental designs range from simple, such as a complete
randomized design (CRD) to more complex block designs with several
factors. In soil science, particularly when analyzing the biological
component of soil, other statistical methods need to be used in order to
analyze soil microbial community diversity and changes in the soil
microbial community which may result because of some external influence
(e.g. temperature, time, fertilizer application, pesticide
application).
A common problem is to
gather environmental or ecological data without having considered the
experimental design. In such instance, it is pertinent that
researchers in environmental sciences have a good background in statistics
including experimental design and data analysis. This is where a
statistician trained in environmetrics can be of great assistance to the
environmental scientist. It is however, important that the
environmetrician has a good understanding of environmental sciences
including specialized applications, for example split-plot designs, which
are common in agricultural trials. As such, sufficient knowledge of
the statistician in the environmental science and that of the
environmental researcher in statistics should make experimental design and
analysis uncomplicated.
|
Richard
Routledge Department of Statistics and Actuarial Science Simon
Fraser University routledg@stat.sfu.ca Title (panel discussion): Impediments to
Environmental Research
A variety of impediments to
environmental research will be discussed, including funding priorities,
societal values, political pressures and special difficulties associated
with doing research on complex environmental interactions.
Title
(multimedia): Rivers Inlet Ecosystem Study: A Case Study in Ecosystem
Research
A major ecosystem study in Rivers Inlet on
the British Columbia Central Coast will be used to illustrate the sort of
statistical challenges that can emerge from such projects. Rivers Inlet is
being used as a model ecosystem for studying the causes of the declines in
many coastal sockeye salmon populations in British Columbia. The study is
focusing on potential food shortages associated with changes to early
spring water conditions and plankton dynamics. Exploratory statistical
analyses of historic time series demonstrate poor survival for
outmigrating juvenile sockeye salmon in years when early spring runoff is
unsually high. To gain further insight, a team of scientists has been
assembled to study the hydrodynamics, plankton dynamics, and fish
migration behaviour and feeding biology. Statistical challenges include,
e.g., comparisons of highly stochastic tracks of drifters (free-floating
objects designed to track surface water movements) to predictions from a
deterministic hydrodynamic model.
|
David
Stanford Department of Statistical and Actuarial Sciences University
of Western Ontario stanford@stats.uwo.ca Title: Analyzing Extra-Fire-Fighting Costs in
the Province of Ontario
This work is in partnership with
A. Ian McLeod and Ou Feng of UWO, and Den Boychuk and Al Tithecott of MNR.
|
Román
Viveros
Department of Mathematics and Statistics
McMaster
University
rviveros@math.mcmaster.ca
(discussant: no
abstract)
|