Philip Jonathan

Discriminant Analysis with Singular Covariance Matrices: Methods and Applications to Spectroscopic Data

SUMMARY Currently popular techniques such as experimental spectroscopy and computer-aided molecular modelling lead to data having very many variables observed on each of relatively few individuals. A common objective is discrimination between two or more groups, but the direct application of standard discriminant methodology fails because of singularity of covariance matrices. The problem has been circumvented in the past by prior selection of a few transformed variables, using either principal component analysis or partial least squares. Although such selection ensures nonsingularity of matrices, the decision process is arbitrary and valuable information on group structure may be lost. We therefore consider some ways of estimating linear discriminant functions without such prior selection. Several spectroscopic data sets are analysed with each method, and questions of bias of assessment procedures are investigated. All proposed methods seem worthy of consideration in practice.

On the use of cross-validation to assess performance in multivariate prediction

We describe a Monte Carlo investigation of a number of variants of cross-validation for the assessment of performance of predictive models, including different values of k in leave-k-out cross-validation, and implementation either in a one-deep or a two-deep fashion. We assume an underlying linear model that is being fitted using either ridge regression or partial least squares, and vary a number of design factors such as sample size n relative to number of variables p, and error variance. The investigation encompasses both the non-singular (i.e. n > p) and the singular (i.e. n ≤ p) cases. The latter is now common in areas such as chemometrics but has as yet received little rigorous investigation. Results of the experiments enable us to reach some definite conclusions and to make some practical recommendations.

Modeling the Seasonality of Extreme Waves in the Gulf of Mexico

Statistics of storm peaks over threshold depend typically on a number of covariates including location, season, and storm direction. Here, a nonhomogeneous Poisson model is adopted to characterize storm peak events with respect to season for two Gulf of Mexico locations. The behavior of storm peak significant wave height over threshold is characterized using a generalized Pareto model, the parameters of which vary smoothly with season using a Fourier form. The rate of occurrence of storm peaks is also modeled using a Poisson model with rate varying with season. A seasonally varying extreme value threshold is estimated independently. The degree of smoothness of extreme value shape and scale and the Poisson rate with season are regulated by roughness-penalized maximum likelihood; the optimal value of roughness is selected by cross validation. Despite the fact that only the peak significant wave height event for each storm is used for modeling, the influence of the whole period of a storm on design extremes for any seasonal interval is modeled using the concept of storm dissipation, providing a consistent means to estimate design criteria for arbitrary seasonal intervals. The characteristics of the 100 year storm peak significant wave height, estimated using the seasonal model, are examined and compared with those estimated ignoring seasonality.

Uncertainties in Extreme Wave Height Estimates for Hurricane-Dominated Regions

Inherent uncertainties in estimation of extreme wave heights in hurricane-dominated regions are explored using data from the GOMOS Gulf of Mexico hindcast for 1900-2005. In particular, the effect of combining correlated values from a neighborhood of 72 grid locations on extreme wave height estimation is quantified. We show that, based on small data samples, extreme wave heights are underestimated and site averaging usually improves estimates. We present a bootstrapping approach to evaluate uncertainty in extreme wave height estimates. We also argue in favor of modeling supplementary indicators for extreme wave characteristics, such as a high percentile (95%) of the distribution of 100-year significant wave height, in addition to its most probable value, especially for environments where the distribution of 100-year significant wave height is strongly skewed.

Locating and quantifying gas emission sources using remotely obtained concentration data

We describe a method for detecting, locating and quantifying sources of gas emissions to the atmosphere using remotely obtained gas concentration data; the method is applicable to gases of environmental concern. We demonstrate its performance using methane data collected from aircraft. Atmospheric point concentration measurements are modelled as the sum of a spatially and temporally smooth atmospheric background concentration, augmented by concentrations due to local sources. We model source emission rates with a Gaussian mixture model and use a Markov random eld to represent the atmospheric background concentration component of the measurements. A Gaussian plume atmospheric eddy dispersion model represents gas dispersion between sources and measurement locations. Initial point estimates of background concentrations and source emission rates are obtained using mixed‘2-‘1 optimisation over a discretised grid of potential source locations. Subsequent reversible jump Markov chain Monte Carlo inference provides estimated values and uncertainties for the number, emission rates and locations of sources unconstrained by a grid. Source area, atmospheric background concentrations and other model parameters, including plume model spreading and Lagrangian turbulence time scale, are also estimated. We investigate the performance of the approach rst using a synthetic problem, then apply the method to real airborne data from a 1600km 2 area containing two landlls,

Discriminant analysis with singular covariance matrices. A method incorporating cross-validation and efficient randomized permutation tests

A computationally efficient approach has been developed to perform two‐group linear discriminant analysis using high‐dimensional data. The analysis is based on Fishers method and incorporates two important validation stages: 1, full leave‐one‐observation‐out cross‐validation; 2, randomized permutation distribution testing. The resulting algorithm and software are known as CREDIT (cross‐validated random‐permutation‐tested efficient discrimination based on an adjusted generalized inverse for the sample total covariance matrix).

A wavelet-based approach for detecting changes in second order structure within nonstationary time series

This article proposes a test to detect changes in general autocovariance structure in nonstationary time series. Our approach is founded on the locally stationary wavelet (LSW) process model for time series which has previously been used for classification and segmentation of time series. Using this framework we form a likelihood-based hypothesis test and demonstrate its performance against existing methods on various simulated examples as well as applying it to a problem arising from ocean engineering.

On the Estimation of Ocean Engineering Design Contours

Understanding extreme ocean environments and their interaction with fixed and floating structures is critical for offshore and coastal design. Design contours are useful to describe the joint behavior of environmental, structural loading, and response variables. We compare different forms of design contours, using theory and simulation, and present a new method for joint estimation of contours of constant exceedance probability for a general set of variables. The method is based on a conditional extremes model from the statistics literature, motivated by asymptotic considerations. We simulate under the conditional extremes model to estimate contours of constant exceedance probability. We also use the estimated conditional extremes model to estimate other forms of design contours, including those based on the first-order reliability method (FORM), without needing to specify the functional forms of conditional dependence between variables. We demonstrate the application of new method in estimation of contours of constant exceedance probability using measured and hindcast data from the Northern North Sea, the Gulf of Mexico, and the North West Shelf of Australia, and quantify their uncertainties using a bootstrap analysis.

Estimation of Storm Peak and Intrastorm Directional–Seasonal Design Conditions in the North Sea

Specification of realistic environmental design conditions for marine structures is of fundamental importance to their reliability over time. Design conditions for extreme waves and storm severities are typically estimated by extreme value analysis of time series of measured or hindcast significant wave height, HS. This analysis is complicated by two effects. Firstly, HS exhibits temporal dependence. Secondly, the characteristics of H sp S are non-stationary with respect to multiple covariates, particularly wave direction and season. We develop directional-seasonal design values for storm peak significant wave height (H sp S ) by estimation of, and simulation under a non-stationary extreme value model for H sp S . Design values for significant wave height (HS) are estimated by simulating storm trajectories of HS consistent with the simulated storm peak events. Design distributions for individual maximum wave height (Hmax) are estimated by marginalisation using the known conditional distribution for Hmax given HS. Particular attention is paid to the assessment of model bias and quantification of model parameter and design value uncertainty using bootstrap resampling. We also outline existing work on extension to estimation of maximum crest elevation and total extreme water level.

Extreme-Value Graphical Models With Multiple Covariates

To assess the risk of extreme events such as hurricanes, earthquakes, and floods, it is crucial to develop accurate extreme-value statistical models. Extreme events often display heterogeneity (i.e., nonstationarity), varying continuously with a number of covariates. Previous studies have suggested that models considering covariate effects lead to reliable estimates of extreme events distributions. In this paper, we develop a novel statistical model to incorporate the effects of multiple covariates. Specifically, we analyze as an example the extreme sea states in the Gulf of Mexico, where the distribution of extreme wave heights changes systematically with location and storm direction. In the proposed model, the block maximum at each location and sector of wind direction are assumed to follow the Generalized Extreme Value (GEV) distribution. The GEV parameters are coupled across the spatio-directional domain through a graphical model, in particular, a three-dimensional (3D) thin-membrane model. Efficient learning and inference algorithms are developed based on the special characteristics of the thin-membrane model. We further show how to extend the model to incorporate an arbitrary number of covariates in a straightforward manner. Numerical results for both synthetic and real data indicate that the proposed model can accurately describe marginal behaviors of extreme events.