SPATIAL STATISTICS FOR
ENVIRONMENTAL
AND ENERGY CHALLENGES
March 8 - 12, 2014
Engineering Science Hall (building 9),
Lecture Hall II, room 2325
ϕ
ω
ωλπ ∞
∇
Ωπ
γ
stsda.kaust.edu.sa
ϕϑ ∏
ϑ √μ
Χ
∑
∇π Ω
SPATIAL STATISTICS FORENVIRONMENTAL
AND ENERGY CHALLENGES
March 8 - 12, 2014
Engineering Science Hall (building 9),
Lecture Hall II, room 2325
SPATIAL STATISTICS FORENVIRONMENTAL
AND ENERGY CHALLENGES
March 8 - 12, 2014
Engineering Science Hall (building 9),
Lecture Hall II, room 2325
Day 1 – March 8
TIME
8:15 - 9:00 a.m.
EVENT
Breakfast
9:00 - 10:30 a.m.
Tutorial 1A:
Spatial Statistics for Environmental Challenges
Prof. Douglas Nychka (NCAR)
10:30 - 11:00 a.m.
Break
11:00 a.m. - 12:30 p.m.
12:30 - 1:30 p.m.
Tutorial1B:
Spatial Statistics for Environmental Challenges
Prof. Douglas Nychka (NCAR)
Lunch in Campus Diner
1:30 - 3:00 p.m.
Tutorial 2A:
Spatial Statistics for Energy Challenges
Prof. Tilmann Gneiting (HITS and KIT)
3:00 - 3:30 p.m.
Break
3:30 - 5:00 p.m.
Tutorial 2B: Spatial Statistics for Energy Challenges
Prof. Tilmann Gneiting (HITS and KIT)
6:00 - 9:00 p.m.
Dinner at KAUST Beach
SPATIAL STATISTICS FORENVIRONMENTAL
AND ENERGY CHALLENGES
March 8 - 12, 2014
Engineering Science Hall (building 9),
Lecture Hall II, room 2325
Day 2 – March 9
TIME
EVENT
8:15 – 8:45 a.m.
Breakfast
8:45 - 9:00 a.m.
Opening: Dean Mootaz Elnozahy (KAUST)
9:00 - 9:45 a.m.
A non-parametric entropy based approach to detect changes in
climate extremes
Prof. Philippe Naveau (LNRS, CNRS)
9:45 - 10:30 a.m.
Simulating the regional climate in Middle East climate using
high-resolution general circulation and nested limited area models
Prof. Georgiy Stenchikov (KAUST)
10:30 - 11:00 a.m.
Break
11:00 - 11:45 a.m.
Assessing fit in Bayesian models for spatial processes
Prof. Mikyoung Jun (Texas A&M University)
11:45 a.m. - 12:30 p.m.
12:30 - 2:00 p.m.
Non-parametric cross covariograms for multivariate spatial data
Prof. Hao Zhang (Purdue University)
Lunch at the Campus Diner
2:00 - 2:45 p.m.
A stochastic space-time model for intermittent precipitation
occurrences
Prof. Ying Sun (Ohio State University)
2:45 - 3:30 p.m.
The need for improved assessment, evaluation and integration
techniques in hydrological science
Prof. Matthew McCabe (KAUST)
3:30 - 4:00 p.m.
Break
4:00 - 4:45 p.m.
Assimilation of reservoir production data by
the Ensemble Kalman Filter
Prof. Henning Omre (Norwegian University of Science and Technology)
4:45 - 5:30 p.m.
Space-time mapping of ocean chlorophyll:
Dynamical or data driven problem
Prof. Ibrahim Hoteit (KAUST)
7:00 - 9:00 p.m.
Dinner: BBQ at the Island Recreation Center
SPATIAL STATISTICS FORENVIRONMENTAL
AND ENERGY CHALLENGES
March 8 - 12, 2014
Engineering Science Hall (building 9),
Lecture Hall II, room 2325
Day 3 – March 10
TIME
EVENT
8:15 - 9:00 a.m.
Breakfast
9:00 - 9:45 a.m.
State-of-the-art in probabilistic forecasting of wind and
solar power generation
Prof. Henrik Madsen (Technical University of Denmark)
9:45 - 10:30 a.m.
Assessing the performance of model-based clustering methods in
multivariate time series with application to identifying
regional wind regimes
Prof. Amanda Hering (Colorado School of Mines)
10:30 - 11:00 a.m.
Break
11:00 - 11:45 a.m.
Multivariate max-stable spatial processes
Prof. Simone Padoan (Bocconi University of Milan)
11:45 a.m. - 12:30 p.m.
A fresh look at Stein’s half spectral spatial-temporal model
Prof. John Kent (University of Leeds)
12:30-2:00 p.m.
Poster Session lunch in Library
2:00 - 2:45 p.m.
Spatial regression with PDE regularization
Prof. Laura Sangalli (Politecnico di Milano)
2:45 - 3:30 p.m.
Multiscale modeling of wear degradation in cylinder liner
Prof. Raul Tempone (KAUST)
3:30 - 4:00 p.m.
Break
4:00 - 5:30 p.m.
Visit of CORNEA and SHAHEEN for external participants
7:00 - 9:00 p.m.
Dinner at Al-Marsa Restaurant
SPATIAL STATISTICS FORENVIRONMENTAL
AND ENERGY CHALLENGES
March 8 - 12, 2014
Engineering Science Hall (building 9),
Lecture Hall II, room 2325
Day 4 – March 11
TIME
EVENT
8:15 - 9:00 a.m.
Breakfast
9:00 - 9:45 a.m.
Equivalent kriging
Prof. William Kleiber (University of Colorado at Boulder)
9:45 - 10:30 a.m.
Extending tapering to multivariate spatial processes:
concepts and illustrations
Prof. Reinhard Furrer (University of Zuerich)
10:30 - 11:00 a.m.
Break
11:00 - 11:45 a.m.
Nonseparable and stationary covariance functions
on spheres cross time
Prof. Emilio Porcu (University Federico Santa Maria)
11:45 a.m. - 12:30 p.m.
12:30 - 2:00 p.m.
New classes of nonseparable space-time covariance functions
Prof. Tatiyana Apanasovich (George Washington University)
Lunch at the Campus Diner
2:00 - 2:45 p.m.
Quasi-likelihood for spatial point processes
Prof. Rasmus Waagepetersen (Aalborg University)
2:45 - 3:30 p.m.
Climate change and variability:
A spatio-temporal data mining perspective
Dr. James Faghmous (University of Minnesota)
3:30 - 4:00 p.m.
Break
4:00 - 4:45 p.m.
Spatial statistics and coral reef ecology
Prof. Michael Berumen (KAUST)
4:45-5:00 p.m.
Closing
Day 5 – March 12
EVENT
Local activities
Workshop abstracts
and speaker Biographies
Prof. Douglas Nychka
NCAR
Douglas Nychka is a statistical scientist
with an interest in the problems posed
by geophysical data sets. His PhD (1983)
is from the University of Wisconsin (US),
and he subsequently spent 14 years as a
faculty member at North Carolina State
University (US). His research background
in fitting curves and surfaces lead to
an interest in the analysis of spatial
and environmental data. In 1997 he
assumed leadership of the Geophysical
Statistics Project at the National Center
for Atmospheric Research (NCAR; US),
an NSF program to build collaborative
research and training between statistics
and the geosciences. In 2004 he became
Director of the Institute of Mathematics
Applied to Geosciences, an interdisciplinary
component at NCAR with a focus on
transferring innovative mathematical
models and tools to the geosciences. His
current interests are in quantifying the
uncertainty of numerical experiments that
simulate the Earth's present and possible
future climate and spatial statistics applied
to large data sets. He has received the
Jerry Sacks Award for Multidisciplinary
Research (2004) and is a Fellow of the
American Statistical Association.
Short course on Spatial
Statistics for Environmental
Challenges: Our physical world:
Finding the curves and surfaces
This series of lectures will be an introduction to the statistics of estimating
smooth functions from observations and simulations of the environment.
The applications of these methods are ubiquitous: from spatial statistics
for temperature observations, to inverting remote sensed measurements, to
summarizing the complex simulations from Earth system models. In all of these
areas, the basic problem is finding a curve or surface in the midst of noisy and
irregular data, and, once found, quantifying the uncertainty in the estimate.
The key is to distill the problem into two parts: a statistical model that describes
how the observations are related to unknown function and another model for
the unknown function itself. Using maximum likelihood or Bayes theorem one
can use these parts to estimate the function. One useful connection is the
equivalence of these statistical techniques with data smoothers and variational
methods such as splines. Examples will be given using digital elevation models,
extremes, and paleoclimate.
Short course on Spatial
Statistics for Energy Challenges
I will discuss the correlation theory of stochastic processes on Euclidean
domains and spheres, which offers a wide range of challenging open problems.
Applications and case studies in weather, climate, and energy research call for
an increased involvement of probabilists and statisticians.
Prof. Tilmann Gneiting
HITS and KIT
Tilmann Gneiting is Group Leader,
Heidelberg Institute of Theoretical
Studies (HITS; Germany), and Professor
of Computational Statistics, Karlsruhe
Institute of Technology (KIT; Germany).
Previously, he held faculty positions at
Heidelberg University (Germany), and
at the University of Washington (US).
Prof. Tilmann's research focuses on the
theory and practice of forecasting, and
spatial and spatio-temporal statistics,
with applications to meteorological,
hydrologic, and economic problems,
among others. His work on probabilistic
forecasting is supported by an Advanced
Grant from the European Research
Council. Tilmann also serves at Editor
for Physical Science, Computing,
Engineering, and the Environment at
The Annals of Applied Statistics.
Prof. Philippe Naveau
Laboratoire des Sciences du Climat et
l’Environnement
After obtaining his PhD in Statistics
at Colorado State University (US)
in 1998, Dr. Philippe Naveau was a
visiting Scientist at National Center
for Atmospheric Research in Boulder,
Colorado (US) for three years. Then
he was an assistant professor in the
Applied Math Department of Colorado
University (US; 20022004-). Since 2004,
he has been a research scientist at the
French National Research Center (CNRS),
and his research work has focused on
environmental statistics, especially in
analyzing extremes events.
A non-parametric entropy based
approach to detect changes in
climate extremes
In this talk, our goal is to provide and study a non-parametric estimator of the
divergence for large excesses. Fundamental features in an extreme value analysis
are captured by the tail behavior. In its original form, the divergence is not
expressed in terms of tails but in function of probability densities. One important
aspect of this work is to propose an approximation of the divergence in terms of
the tail distributions. This leads to a new non-parametric divergence estimator
tailored for excesses. Its properties are studied. This application focuses primarily
on temperature extremes measured at 24 European stations with at least 90
years of data. Here the term extremes refers to rare excesses of daily maxima and
minima. As we do not want to hypothesize any parametric form of such possible
changes, we propose a new non-parametric estimator based on the KullbackLeibler divergence tailored for extreme events. Our approach is also applied to
seasonal extremes of daily maxima and minima for our 24 selected stations (joint
work with A. Guillou and T. Riestch).
Simulating the regional
climate in Middle East climate
using high-resolution general
circulation and nested limited
area models
The arid and semi-arid regional climates of the Middle Eastern and North
African dry subtropics are closely linked to the large-scale Hadley Circulation
and Monsoon Systems and therefore could not be completely described using
the conventional nested model approach. Located in the heart of the dust belt,
this region experiences tremendous aerosol radiative forcing that has to be better
quantified to reliably describe current climate and predict future climate change.
In my talk I will present recent results, obtained in my group, on calculating
regional climate in the Middle East using global 25-km resolution atmospheric
GCM and simulating dust effects and dust storms using a nested regional model
fully coupled with an interactive dust module.
Prof. Georgiy Stenchikov
King Abdullah University of Science and
Technology (KAUST)
Dr. Georgiy Stenchikov is Professor in
the Division of Physical Sciences and
Engineering and Chair of the Earth
Sciences and Engineering Program at
King Abdullah University of Science and
Technology (KAUST). He graduated from
the Moscow Physical Technical Institute
(Russia) in 1973, where he also obtained
his PhD and habilitation in 1977 and
1989, respectively. He began his research
career in the Computer Center of the USSR
Academy of Sciences, the top Russian
numerical modeling research institute,
where he focused on numerical methods,
radiation gas dynamics, and absorption
of laser radiation in plasmas for laser
fusion applications. In the 1980s Prof.
Stenchikov turned his attention to climate
modeling. From 1986 to 1992, he led a
research group in the Computer Center
that studied anthropogenic impacts on
the Earth’s climate and environmental
systems. In 1992 Prof. Stenchikov moved
to the United States, where he joined the
faculty at the University of Maryland and
then at Rutgers University. He conducted
interdisciplinary studies in the broad field
of climate modeling, atmospheric physics,
and environmental sciences, and published
on the effects of severe thunderstorms
on chemical balances in the troposphere,
stratosphere-troposphere exchange, aerosol
radiative forcing, stretched-grid general
circulation modeling, climate downscaling
using regional models, and impacts of
explosive volcanic eruptions on the climate.
Prof. Mikyoung Jun
Texas A&M University
Mikyoung Jun obtained her PhD in
Statistics from the University of Chicago
(US) in 2005. She was an Assistant
Professor at Texas A&M University,
Statistics, from 2005 – 2012, and now
is an Associate Professor at Texas A&M,
Statistics, since 2012.
Assessing fit in Bayesian models
for spatial processes
Gaussian random fields are frequently used to model spatial and spatialtemporal data, particularly in geostatistical settings. As much of the
attention of the statistics community has been focused on defining and
estimating the mean and covariance functions of these processes, little
effort has been devoted to developing goodness-of-fit tests to allow
users to assess the models’ adequacy. We describe a general goodness-offit test and related graphical diagnostics for assessing the fit of Bayesian
Gaussian process models using pivotal discrepancy measures. Our method
is applicable for both regularly and irregularly spaced observation locations
on planar and spherical domains. The essential idea behind our method is to
evaluate pivotal quantities defined for a realization of a Gaussian random
field at parameter values drawn from the posterior distribution. Because the
nominal distribution of the resulting pivotal discrepancy measures is known,
it is possible to quantitatively assess model fit directly from the output of
MCMC algorithms used to sample from the posterior distribution on the
parameter space. We illustrate our method in a simulation study and in two
applications. This is joint work with Valen Johnson and Matthias Katzfuss
(Texas A&M) and Jianhua Hu (MD Anderson Cancer Center).
Non-parametric cross
covariograms for multivariate
spatial data
For multivariate spatial data, cross covariogram describes the correlation between
difference variables and therefore plays a central role in co-kriging. Current
practice is to build a multivariate covariogram jointly for both the marginal and
cross covariograms and a few such new parametric models have been developed
in recent years. However, parameters in the multivariate covariogram must
satisfy certain constraints in order to yield positive definite covariance matrix.
Consequently, estimation of the multivariate covariogram is often complex. In
addition, it has been reported that the parametric multivariate covariogram may
not yield better prediction results than kriging. In this talk, I propose first to model
each individual marginal covariogram using only marginal data and build the
non-parametric cross covariograms in such a way that the likelihood function
or the predictive scores will be improved. One advantage of this approach is that
it can deal with any marginal covariograms. I will show that the linear model of
coregionalization is a special case of our model. I will discuss numerical algorithms
and demonstrate its performance through both simulated and observed data sets.
Prof. Hao Zhang
Purdue University
Hao Zhang graduated from Peking
University (China) with BS in Mathematics
(1986) and an MS in Statistics (1989). In
1990 he came to the US and earned a PhD
in Statistics from Michigan State University
(1995). Since then, he joined the faculty
at Marquette University (US), Washington
State University (US), and Purdue University
(US), where he has a joint appointment
in Statistics and Forestry and Natural
Resources, and is Associate Head of the
Department of Statistics. He is a Fellow
of American Statistical Association and
an Elected Member of the International
Statistical Institute. His research interests
are focused on modeling spatial and spatiotemporal data that arise in environmental
and climate studies. He has developed
theoretical results and computational
methods for analyzing massive spatial data.
Prof. Ying Sun
Ohio State University
Ying Sun is an assistant professor in the
Department of Statistics at Ohio State
University (US). She received her PhD
degree in Statistics from Texas A&M
University in 2011. Before joining Ohio State
University in 2013, she was a postdoctoral
researcher at the University of Chicago (US;
20122013-) in the research network for
Statistical Methods for Atmospheric and
Oceanic Sciences (US; STATMOS), and at
the Statistical and Applied Mathematical
Sciences Institute (US; SAMSI) in the
Uncertainty Quantification program (20112012). She demonstrated excellence in
research and teaching, published research
papers in top statistical journals as well as
subject matter journals, won multiple best
paper awards from the American Statistical
Association and the Transportation
Research Board National Academies, and
best poster awards from the Workshop
on Environmetrics and the Conference
on Resampling Methods and Highdimensional Data. Her research interests
include spatial and spatio-temporal
statistics, computational methods for large
datasets, numerical model uncertainty
quantification, complex data visualization,
statistics of extremes, time series analysis,
and applications on data and modeldriven scientific challenges from science
and technology.
A stochastic space-time model
for intermittent precipitation
occurrences
Modeling a precipitation field is challenging due to its intermittent and
highly scale-dependent nature. Motivated by the features of high-frequency
precipitation data from a network of rain gauges, we propose a threshold spacetime t random field (tRF) model for 15-minute precipitation occurrences. This
model is constructed through a space-time Gaussian random field (GRF) with
random scaling varying along time or space and time. It can be viewed as a
generalization of the purely spatial tRF, and has a hierarchical representation
that allows for Bayesian interpretation. The randomness of the scaling process
increases the variability across realizations from the GRF, and is shown to capture
the variability of the precipitation occurrence better than the threshold GRF
model for the data we have considered. For model comparisons and diagnostics,
we focus on evaluating whether models can produce the observed conditional
dry and rain probabilities given the neighboring sites have rain or no rain. The
conditional probabilities, along with the marginal rainfall probabilities, are used
to summarize the variability of the precipitation occurrence in space and time,
and useful graphical tools are developed for visualization purpose. Model fitting
and validation are conducted by Monte Carlo simulation-based approaches,
where the statistical efficiency is presented by visualizing a set of the conditional
probabilities calculated from simulations of the fitted models.
The need for improved
assessment, evaluation and
integration techniques in
hydrological sciences
The hydrological sciences represent a broad school of research specializations.
Whether seeking to describe streamflow, precipitation, or evaporation, a wide
variety of approaches are used to characterize the various processes that
make up this discipline. These might include empirical representations, semidistributed approaches or more sophisticated fully-distributed and physically
based descriptions. While model development continues to play a significant
role in hydrology, many of the research challenges lie in the area of robust
parameter estimation, uncertainty quantification, efficient data integration and
comprehensive model evaluation. Here we will review a few of these problems
and the need for new (or existing) statistical techniques to address them. In
discussing these ideas, a particular focus will be directed towards the concept
of robust model evaluation and how spatially distributed information – such as
routinely available from satellite based sensors – might be best integrated or used
to independently evaluate model simulations. The spatial and temporally rich
information content available from space-based systems provides an opportunity
to integrate not just the quantity being measured, but perhaps more importantly,
represent the observed pattern and texture that model systems should be able to
reproduce. To date, the discipline as a whole has done this type of comprehensive
model assessment very poorly.
Prof. Matthew McCabe
King Abdullah University of Science and
Technology (KAUST)
Prof. McCabe studied Civil and
Environmental Engineering at the
University of Newcastle (Australia), where
he received his undergraduate and postgraduate degrees. He then spent five years
in the United States, working at both
Princeton University and then at Los Alamos
National Laboratory in New Mexico. He
returned to Australia in 2008 to take up a
faculty position at the University of New
South Wales, and in late 2012, he joined the
faculty at KAUST. Prof. McCabe's research
interests encompass the modelling and
observation of terrestrial water and energy
cycles. This involves using satellite-based
remote sensing approaches and in-situ
techniques to characterize the terrestrial
and atmospheric systems, with a focus on
describing hydrological behavior across local,
regional, and global scales. These efforts
also involve exploring the effective use and
integration of observations within modeling
systems, with the aim of improving our
capacity to understand (and predict) water
and energy cycle processes.
Prof. Henning Omre
Norwegian University of Science and
Technology
Henning Omre is Professor of Statistics
in the Department of Mathematical
Sciences, Norwegian University of Science
& Technology (NTNU; Norway). He received
his MSc in Statistics at NTNU in 1975, and
received his Ph.D. in Geostatistics from
Stanford University, California (US), in 1985.
He was employed at NR, Oslo (Norway), from
197699-; founded the SAND-group; and has
been a Professor at NTNU since 1992.
Assimilation of reservoir
production data by the
Ensemble Kalman Filter
The ensemble Kalman filter (EnKF) provides an approximate Monte Carlo solution
to forecasts in hidden Markov chain models. The EnKF algorithm will be presented
and discussed. Under Gauss-linear assumptions the EnKF is asymptotically correct
as the number of ensemble members tends towards infinity. For finite ensemble
filters many complications are reported – even under Gauss-linear model
assumptions. Some finite sample results that shed light on these complications
will be presented – with focus on uncertainty quantification. Efficient depletion of
petroleum reservoirs requires reliable descriptions for the reservoir geometry and
hydrocarbon filling. During depletion the hydrocarbon production is monitored,
and this production history must be assimilated into the reservoir description.
This spatio-temporal modelling challenge can be addressed by an EnKF approach.
A case study of production history conditioning of a North Sea reservoir will be
briefly presented.
Space-time mapping of ocean
chlorophyll: Dynamical or data
driven problem
Microscopic marine plants (phytoplankton) form the base of the marine food
chain and play a key role in the climate system. Mapping and understanding
their spatio-temporal variability is therefore an important goal of the present
day oceanography. Phytoplankton contains chlorophyll, a green pigment used
during photosynthesis. With satellite sensors, it is possible to measure chlorophyll
concentration, which is then used as a proxy to indicate the distribution and
amount of phytoplankton. Dynamical and data driven methods could be used
for mapping satellite chlorophyll concentration. Dynamical methods rely on
the availability of a marine ecosystem model, which couples a physical model
with a biological model. The model acts as an interpolator inferring space-time
statistics of the distribution of the chlorophyll concentration from the dynamics,
which get updated by the data. Data-driven statistical methods may provide
an alternative to avoid resorting to complex and costly ecosystem models.
Estimating a space-time covariance model for chlorophyll concentration is not
a trivial task, and may require information about other physical and biological
parameters that cannot be always directly or fully observed. This talk will present
and discuss both approaches and show results from realistic applications in the
Mediterranean Sea and the Red Sea.
Prof. Ibrahim Hoteit
King Abdullah University of Science and
Technology (KAUST)
Ibrahim Hoteit is an Associate Professor
leading the earth fluid modeling and
prediction group at KAUST. He received
his PhD in Applied Mathematics in 2001
from the University of Grenoble (France).
He worked as a research scientist at
Scripps Institute of Oceanography (SIO
– UCSD; US), within the Estimating of
the Circulation and the Climate of the
Ocean (ECCO – MIT/SIO/JPL) Consortium.
His research focuses on the effective use
and integration of dynamical models and
observations to simulate, study, and predict
geophysical fluid systems. This involves
developing and implementing numerical
models and data inversion, assimilation
and uncertainty quantification techniques
suitable for large scale applications.
Prof. Henrik Madsen
Technical University of Denmark
Henrik Madsen has been a professor
in Math. Statistics at the Technical
University of Denmark since 1999. He is
currently Section Head for the Dynamical
Systems Section, and he is leading the
Coordination Committee for Research
on Intelligent Energy Systems at DTU.
He is the author of about 450 scientific
papers. He is a co-author of the new book
Integrating Renewables on the Electricity
Markets (Springer, 2013). His recent books
on statistics are: Time Series Analysis
(Chapman & Hall, 2008) and Introduction
to General and Generalized Linear Models
(Chapman & Hall, 2011).
State-of-the-art in probabilistic
forecasting of wind and solar
power generation
This talk describes state-of-the-art methods for probabilistic forecasting of wind
and solar power. In Denmark, on average of 55% of the total electricity load was
covered by wind power in December 2013, and the key to enable a successfully
integration of such a high share of fluctuating renewable power is to take
advantage of reliable methods for probabilistic forecasting in the operational
dispatch (stochastic optimization) of power generation and stochastic control
techniques. The talk briefly describes the nonlinearities and nonstationarities that
must be taken into account even for generating suitable point predictions. Methods
for taking spatio-temporal dependencies into account are also outlined. However,
the talk focuses on methods embedded in the state-of-the-art methodologies for
probabilistic forecasting of wind and solar power. These methods include adaptive
quantile regression and use of nonlinear stochastic differential equations.
Assessing the performance of
model-based clustering methods
in multivariate time series with
application to identifying
regional wind regimes
Distinct wind conditions driven by prevailing weather patterns exist in every
region around the globe. Knowledge of these conditions can be used to select
and place turbines within a wind project, design controls, and build space-time
models for wind forecasting. Identifying regimes quantitatively and comparing
the performance of different regime identification methods are the goals of
this research. The ability of statistical clustering techniques to correctly assign
hourly observations to a particular regime and to select the correct number of
regimes is studied through simulation. Pressure and the horizontal and vertical
wind components are simulated under two different regimes with a first-order
Markov-switching vector autoregressive model, and the following five clustering
algorithms are applied: (1) classification based on wind direction, (2) k-means, (3)
a nonparametric mixture model, and (4,5) a Gaussian mixture model (GMM) with
one of two covariance structures. The GMM with an unconstrained covariance
matrix has the lowest misclassification rate and the highest proportion of
instances in which two regimes are selected. This method is applied to one year of
averaged hourly wind data observed at twenty meteorological stations. The lagged
wind speed correlations between neighboring sites under upwind and downwind
regimes are shown to differ substantially, and forecasting models based on local
regimes are built and evaluated.
Prof. Amanda Hering
Colorado School of Mines
Amanda S. Hering earned her B.S. in
Mathematics from Baylor University in
1999 and an MS in Statistics from Montana
State University in 2002. She received
her PhD in Statistics from Texas A&M
University in 2009 under the advisement
of Marc G. Genton, where she studied wind
modeling and evaluation, and hypothesis
tests for spatial prediction comparison.
Since completing her PhD, she has been an
assistant professor at Colorado School of
Mines (US) in the Department of Applied
Mathematics and Statistics. She continues
to work on applying multivariate and spatial
statistical methods to problems in wind, the
environment, and defense.
Prof. Simone Padoan
Bocconi University of Milano
Simone Padoan has been an assistant
professor at the Bocconi University of
Milan (Italy) since September 2012. He
has carried out research as a post-doc at
Ecole Poytechnique Fédérale de Lausanne
(Switzerland) from February 2008 to
December 2010; as a research fellow at
the University of Bergamo (Italy) from
January 2011 to September 2011; and as a
senior research fellow at the University of
Padua (Italy) from October 2011 to August
2012. He received his PhD in Statistics in
2008 and his Degree in Statistics in 2004
from the University of Padua (Italy). During
his PhD he spent about two years at the
School of Mathematics and Statistics at the
University of New South Wales (Australia).
His main research interests concern the
theory of extreme values, spatial data
analyses, and estimation methods for
complex models based on Bayesian and
likelihood approaches.
Multivariate max-stable
spatial processes
Analysis of spatial extremes is currently based on univariate processes. Maxstable processes allow the spatial dependence of extremes to be modelled and
explicitly quantified, they are therefore widely adopted in applications. For a
better understanding of extreme events of real processes, such as environmental
phenomena, it may be useful to study several spatial variables simultaneously.
To this end we extend some theoretical results and applications of max-stable
processes to the multivariate setting to analyze extreme events of several
variables observed across space. In particular we study the maxima of independent
replicates of multivariate processes, both in the Gaussian and Student-t cases.
Then we define a Poisson process construction in the multivariate setting and
introduce multivariate versions of the Smith Gaussian extreme-value, the
Schlather extremal-Gaussian and extremal-t, and the Brown-Resnick models.
Inferential aspects of those models based on composite likelihoods are developed.
We present results of various Monte Carlo simulations and of an application to
a dataset of summer daily temperature maxima and minima in Oklahoma, USA,
highlighting the utility of working with multivariate models in contrast to the
univariate case. This work is a joint project with Marc G. Genton and Huiyan Sang.
A fresh look at Stein's half-spectral
spatial-temporal model
A stationary spatial-temporal Gaussian model is specified by a covariance
function (depending on spatial and temporal lags) or alternatively in terms of its
Fourier transform in space or time, or both. Stein proposed a semi-parametric
model specified most simply in the half-spectral domain (i.e. spatial lags and
temporal frequencies). In this talk we develop graphically-based regression
procedures to estimate the various parameters in this model. In addition the
possible presence of a "dimple" effect is examined.
Prof. John Kent
University of Leeds
John Kent is a professor in the Department
of Statistics, University of Leeds (UK). He
received his PhD in Statistics from the
University of Cambridge and moved to
Leeds in 1977, where he has spent the
bulk of his academic career. His research
interests include infinite divisibility,
directional and multivariate analysis,
inference, robustness, shape analysis, and
spatial and spatial-temporal modelling.
Prof. Laura Sangalli
Politecnico di Milano
Laura Sangalli is a researcher at MOX
Laboratory for Modeling and Scientific
Computing, Dipartimento di Matematica,
Politecnico di Milano (Italy). Her main
research interests concern the development
of models for the analysis of functional
and spatial data, more complex data
structures, and object data (http://mox.
polimi.it/users/sangalli).
Spatial regression with PDE
regularization
Spatial regression with differential regularization is a novel class of models
for the accurate estimation of surfaces and spatial fields that merges
advanced statistical methodology and scientific computing techniques.
Thanks to the combination of potentialities from these two scientific areas,
the proposed class of models has important advantages with respect to
classical techniques used in spatial data analysis. Spatial regression with
differential regularization is able to efficiently deal with data distributed over
irregularly shaped domains with complex boundaries, strong concavities,
and interior holes. Moreover, it can comply with specific conditions at the
boundaries of the domain, which is fundamental in many applications to
obtain meaningful estimates. The proposed models can also handle data
distributed over general bidimensional Riemannian manifold domains, some
of the few methods existing in literature for this type of data structures.
Moreover, spatial regression with differential regularization has the capacity
to incorporate problem-specific priori information about the spatial structure
of the phenomenon under study, formalized in terms of a governing PDE, and
allows for a very flexible modeling of space variation accounting naturally
for anisotropy and non-stationarity. Space-varying covariate information is
included in the models via a semiparametric framework. The estimators have
a penalized regression form, they are linear in the observed data values, and
have good inferential properties. The use of numerical analysis techniques,
and specifically of finite elements, makes the models computationally very
efficient. The method is illustrated in various applied contexts, including
demographic data and data coming from eco-dopplers and computational
fluid dynamics simulations. This line of research is developed within the FIRB
starting grant project SNAPLE http://mox.polimi.it/users/sangalli/firbSNAPLE.
html The seminar is based on joint work with Laura Azzimonti, Bree Ettinger,
Fabio Nobile, Simona Perotto, Jim Ramsay, and Piercesare Secchi.
Multiscale modeling of wear
degradation in cylinder liner
Every mechanical system is naturally subjected to some kind of wear process
that, at some point, will cause failure in the system if no monitoring or treatment
process is applied. Since failures often lead to high economical costs, it is essential
both to predict and to avoid them. To achieve this, a monitoring system of the wear
level should be implemented to decrease the risk of failure. In this work, we take
a first step into the development of a multiscale indirect inference methodology
for state-dependent Markovian pure jump processes. This allows us to model the
evolution of the wear level, and to identify when the system reaches some critical
level that triggers a maintenance response. Since the likelihood function of a
discretely observed pure jump process does not have an expression that is simple
enough for standard non-sampling optimization methods, we approximate this
likelihood by expressions from upscaled models of the data. We use the Master
Equation to assess the goodness-of-fit and to compute the distribution of the
hitting time to the critical level.
Prof. Raul Tempone
King Abdullah University of Science and
Technology (KAUST)
Author of forty journal articles and
conference publications, Dr. Tempone’s
research interests are in the mathematical
foundation of computational science and
engineering. More specifically, he has
focused on a posteriori error approximation
and related adaptive algorithms for
numerical solutions of various differential
equations, including ordinary differential
equations, partial differential equations,
and stochastic differential equations. He
is also interested in the development and
analysis of efficient numerical methods
for uncertainty quantification and
Bayesian model validation. The areas of
application he considers include, among
others, engineering, chemistry, biology,
and physics, as well as social science and
computational finance.
Prof. William Kleiber
University of Colorado at Boulder
William Kleiber is an assistant professor in
the Department of Applied Mathematics
at the University of Colorado at Boulder
(US). He received his PhD in Statistics
from the University of Washington (US),
and moved to Colorado for a postdoctoral position at the National Center
for Atmospheric Research. His research
focuses on geophysical and environmental
problems, with particular emphasis on
spatial statistics, computer experiments and
stochastic modeling of physical systems.
Equivalent kriging
Most modern spatially indexed datasets are very large, with sizes commonly
ranging from tens of thousands to millions of locations. Spatial analysis often
focuses on spatial smoothing using the geostatistical technique known as
kriging. Kriging requires covariance matrix computations whose complexity
scales with the cube of the number of spatial locations, making analysis
infeasible or impossible with large datasets. We introduce an approach to
kriging in the presence of large datasets called equivalent kriging, which relies
on approximating the kriging weight function using an equivalent kernel.
Resulting kriging calculations are extremely fast and feasible in the presence
of massive spatial datasets. We derive closed form kriging approximations for
multiresolution classes of spatial processes, as well as under any stationary
model, including popular choices such as the Matern. The theoretical justification
for equivalent kriging also leads to a convenient correction term for irregularly
spaced observations. Estimation can proceed by either cross-validation or
generalized cross-validation. Equivalent kriging is illustrated on two simulated
datasets, and a monthly average precipitation dataset whose size prohibits
traditional geostatistical approaches.
Extending tapering to
multivariate spatial processes:
concepts and illustrations
Parameter estimation for, and smoothing or interpolation of, spatially or spatiotemporally correlated random processes is used in many areas and often requires
the solution of a large linear system based on the covariance matrix of the
observations. In recent years the dataset sizes have steadily increased such that
straightforward statistical tools are computationally too expensive to be used. In
the univariate context, tapering, i.e. creating sparse approximate linear systems,
has been shown to be an efficient tool in both the estimation and prediction
setting. In this talk we present a short review of tapering in the context of temporal
and spatial statistics. Key concepts in the framework of estimation and prediction
for univariate spatial processes are discussed. A pragmatic asymptotic setting for
the extension of tapering to multivariate spatial processes is given and illustrated.
We conclude with open problems and challenges of tapering in the context
environmental and energy modeling.
Prof. Reinhard Furrer
University of Zurich
Reinhard Furrer received an undergraduate
degree (diploma) in mathematics of the
Swiss Federal Institute of Technology in
1998 and completed his Ph.D. in Statistics
at the same institute in 2002. In 2002
Dr. Furrer moved to Boulder, Colorado
(US), starting a position as a postdoctoral
fellow in the Geophysical Statistics Project
at the National Center for Atmospheric
Research (US). From 2005 to 2009, he was
an assistant professor in the Mathematics
and Computer Science department at
the Colorado School of Mines in Golden,
Colorado (US). In 2009 Dr. Furrer accepted
a position as an assistant professor at the
Institute of Mathematics at the University
of Zurich (Switzerland), continuing bridging
statistical methodologies with cutting edge
science projects. Keywords describing Dr.
Furrer’s core statistical research interests are
spatial and spatio-temporal statistics, large
(spatial) datasets, non stationary spatial
processes, statistical evaluation of climate
model output, and ensemble Kalman filter.
Many methodological papers are addressing
the modeling and implementation of large
spatial datasets using sparse covariance
models. Recently Dr. Furrer got involved with
a large interdisciplinary collaborative project
entitled Global Change and Biodiversity.
Prof. Emilio Porcu
University Federico Santa Maria
Emilio Porcu was born in Sardinia. He
completed his PhD at the University of
Milan (Italy) and studied geostatistics
at the Ecole des Mines de Paris (France).
He has worked in Italy, Spain, and
Germany, and now he is a professor at the
University Federico Santa Maria (Chile).
His other interests are in: statistics and
stochastic processes; with applications
to atmospheric, environmental and earth
sciences, petroleum engineering, and
economics, among other disciplines;
space–time processes, geostatistics,
point processes; mathematics: positive
definite functions; variograms; completely
monotone functions, locally compact
abelian groups, Laplace transforms, and
potential theory. Recently he has coined
a discipline called “Seismomatics,” which
denotes the fusion of mathematics,
statistics, and data mining for helping
those discipline interested in the analysis
and forecast of natural catastrophes under
the paradigm of random field theory.
Special emphasis is put on analysis and
assessment of earthquakes, tsunamis,
avalanches, flooding, atmospheric
and water pollution, and many other
phenomena of interest for the health of
the human beings and for the economic
health of countries. You can find more info
about him, his research, and his group at
eporcu.mat.utfsm.cl.
Nonseparable and stationary
covariance functions on spheres
cross time
In this talk, we illustrate covariance models associated to Gaussian fields evolving
temporally over spheres. On one hand, we show that celebrated constructions,
proposed in earlier literature on Euclidean spaces, can be readapted for the
case of great circle distance. We then illustrate some models being valid on
spheres cross time but not on Euclidean spaces. We also propose models of
cross-covariances associated to multivariate Gaussian fields. A simulation study
assesses the importance of using the great circle distance instead of other
approximations for both estimation and prediction. We revisit TOMS data with
our model and investigate differences in terms of estimation and prediction
when using different metrics.
New classes of nonseparable
space-time covariance functions
Statistical methods for the analysis of space-time data are of great interest
for many areas of application. Geostatistical approaches to spatiotemporal
estimation and prediction heavily rely on appropriate covariance models. In
my talk I will first give an overview of techniques to build valid space-time
covariances that must satisfy the positive definiteness constraint. Then I will
discuss the specific properties of covariance functions and how they relate
to spatial and temporal marginal processes as well as their interaction. The
highlighted critical aspects to model building will be used to motivate the
proposed family of nonseparable space-time covariance structures which
have the celebrated Matern family for their spatial margins. I will also describe
a simple modification of the new family to address the lack of symmetry. The
application of the proposed methodologies will be illustrated on the datasets
from environmental science and meteorology.
Prof. Tatiyana Apanasovich
George Washington University
Tatiyana Apanasovich’s research
spans the following broad areas and
their intersections: Spatial statistics,
semiparametric/nonparametric regression
methods, measurement error models,
functional data analysis, and statistical
genetics. For her research efforts she has
been awarded NSF and NIH grants.
Prof. Rasmus
Waagepetersen
Aalborg University
Rasmus Waagepetersen obtained his
PhD in Mathematical Statistics at
Aarhus University (Denmark) in 1997.
He subsequently worked as a scientist
at the Danish Institute of Agricultural
Sciences until 2000. Since 2000, he has
been affiliated with Aalborg University
(Denmark), where he became a full
Professor in 2010. In 2008 and 2009 he
was on leave from Aalborg University to
work with credit risk in Spar Nord Bank
(Denmark). His main research interests are
statistics for spatial point processes, mixed
models, and computational methods in
quantitative genetics. Prof. Waagepetersen
is a co-author of the monograph
"Statistical inference and simulation for
spatial point processes."
Quasi-likelihood for spatial
point processes
Fitting regression models for intensity functions of spatial point processes is of
great interest in ecological and epidemiological studies of association between
spatially referenced events and geographical or environmental covariates.
When Cox or cluster process models are used to accommodate clustering not
accounted for by the available covariates, likelihood based inference becomes
computationally cumbersome due to the complicated nature of the likelihood
function and the associated score function. It is therefore of interest to consider
alternative more easily computable estimating functions. We derive the optimal
estimating function in a class of first-order estimating functions. The optimal
estimating function depends on the solution of a certain Fredholm integral
equation which in practice is solved numerically. The derivation of the optimal
estimating function has close similarities to the derivation of quasi-likelihood
for standard data sets. The approximate solution is further equivalent to a quasilikelihood score for binary spatial data. We therefore use the term quasi-likelihood
for our optimal estimating function approach. We demonstrate in a simulation
study and a data example that our quasi-likelihood method for spatial point
processes is both statistically and computationally efficient.
Climate change and variability:
a spatio-temporal data mining
perspective
Our planet is experiencing simultaneous changes in global population, urbanization,
and climate. These changes, along with the rapid growth of climate data and
increasing popularity of data mining techniques may lead to the conclusion that the
time is ripe for data mining to spur major innovations in climate science. However,
climate data bring forth unique challenges that are unfamiliar to the traditional data
mining literature, and unless they are addressed, data mining will not have the same
impact that it has had on fields such as biology or e-commerce. This talk provides a
technical audience with an introduction to mining climate data with an emphasis on
the singular characteristics of the datasets and research questions climate science
attempts to address. We demonstrate some of the concepts discussed in the earlier
parts of the talk with a spatio-temporal pattern mining application to monitor
global ocean eddy dynamics. We show that insightfully mining the spatio-temporal
context of climate datasets can yield significant improvements in the performance
of (non space-time aware) learning algorithms.
Dr. James Faghmous
University of Minnesota
James H. Faghmous obtained his PhD in
Computer Science from the University of
Minnesota - Twin Cities (US). His doctoral
research was part of a five-year $10M NSFfunded Expeditions in Computing grant to
develop novel numerical techniques to study
and monitor climate change. As part of the
Expeditions team, Dr. Faghmous developed
scalable data mining algorithms to analyze
large climate datasets. His research has
been funded by an NIH Neuro-PhysicalComputational Graduate Fellowship, an
NSF Graduate Research Fellowship, an NSF
Nordic Research Opportunity Fellowship,
and a University of Minnesota Doctoral
Dissertation Fellowship. Dr. Faghmous
graduated in 2006 with a BSc in Computer
Science from the City of College of New
York, where he was a Rhodes and a Gates
Scholar nominee.
Prof. Michael Berumen
King Abdullah University of Science and
Technology (KAUST)
Prof. Berumen received a Zoology degree
from the University of Arkansas (US) in 2001.
He then attended James Cook University
(Australia) to pursue graduate studies in
coral reef ecology, specializing in the life
history and ecology of butterflyfishes. He
was awarded his PhD in 2007. Prof. Berumen
accepted a postdoctoral fellowship at the
Woods Hole Oceanographic Institution
(WHOI; US), where he focused on larval
connectivity in coral reef fishes. His research
focuses on advancing our understanding
of Red Sea coral reefs, and more on broadly
on making contributions to movement
ecology, which is a critical aspect of
developing conservation plans in the marine
environment. He is particularly interested in
connectivity questions ranging from larval
dispersal to large distance migrations of
adult fishes.
Spatial statistics and coral
reef ecology
The Coral Reef Ecology Lab (reefecology.kaust.edu.sa) within KAUST's Red Sea
Research Center is the home of several projects addressing large-scale ecological
studies in the Red Sea and beyond. Many of the datasets generated in our work
span spatial scales that introduce some analytical challenges. This data ranges
from rapid ecological community surveys along the entirety of the Saudi Arabian
Red Sea coast to half a decade of observations of whale sharks utilizing a feeding
"hotspot." All of these data can reveal biological patterns or behaviors that can
potentially be used to design more effective conservation measures for various
species. Leveraging expertise in spatial statistics allows for greater impact of
existing data and improved sampling design for future projects.
Notes
Notes
π ϕ∏
ϑ
√μ
ϕ
ω
∞
λπ
ϕϑ ∏
ϑ √μ
∇
Χ
∑
Ω
∇
Χ
∑
Ω
Organizers:
Prof. Marc G. Genton (KAUST, Spatio-Temporal Statistics and Data Analysis)
Prof. Raul Tempone (KAUST, Stochastic Numerics Research and SRI-UQ)
Dr. Fabrizio Ruggeri (CNR-IMATI)