Final published version
Research output: Contribution to Journal/Magazine › Journal article › peer-review
Research output: Contribution to Journal/Magazine › Journal article › peer-review
}
TY - JOUR
T1 - Fast computation of large scale marginal extremes with multi-dimensional covariates
AU - Raghupathi, L.
AU - Randell, D.
AU - Ewans, K.
AU - Jonathan, P.
PY - 2016/3/1
Y1 - 2016/3/1
N2 - Safe and reliable design and operation of fixed and floating marine structures often located in remote and hostile environments is challenging. Rigorous extreme value analysis of meteorological and oceanographic data can greatly aid the design of such structures. Extreme value analysis is typically undertaken for single spatial locations or for small neighbourhoods; moreover, non-stationary effects of covariates on extreme values are typically accommodated in an ad-hoc manner. The objective of the work summarised here is to improve design practice by estimating environmental design conditions (such as return values for extreme waves, winds and currents) for a whole ocean basin, including additional covariate effects (such as storm direction) as necessary, in a consistent manner. Whole-basin non-stationary extreme value modelling is computationally complex, requiring inter-alia the estimation of tail functions, the parameters of which vary with respect to multi-dimensional covariates characterised by us using tensor products of penalised B-splines. We outline two technical contributions which make whole-basin non-stationary analysis feasible. Firstly, we adopt generalised linear array methods to reduce the computational burden of matrix manipulations. Secondly, using high-performance computing, we develop a parallel implementation of maximum likelihood estimation for the generalised Pareto distribution. Together, these innovations allow estimation of rigorous whole-basin extreme value models in reasonable time. We evaluate the new approach in application to marginal extreme value modelling of storm peak significant wave heights in two ocean basins, accommodating spatial and directional covariate effects. © 2015 Elsevier B.V.
AB - Safe and reliable design and operation of fixed and floating marine structures often located in remote and hostile environments is challenging. Rigorous extreme value analysis of meteorological and oceanographic data can greatly aid the design of such structures. Extreme value analysis is typically undertaken for single spatial locations or for small neighbourhoods; moreover, non-stationary effects of covariates on extreme values are typically accommodated in an ad-hoc manner. The objective of the work summarised here is to improve design practice by estimating environmental design conditions (such as return values for extreme waves, winds and currents) for a whole ocean basin, including additional covariate effects (such as storm direction) as necessary, in a consistent manner. Whole-basin non-stationary extreme value modelling is computationally complex, requiring inter-alia the estimation of tail functions, the parameters of which vary with respect to multi-dimensional covariates characterised by us using tensor products of penalised B-splines. We outline two technical contributions which make whole-basin non-stationary analysis feasible. Firstly, we adopt generalised linear array methods to reduce the computational burden of matrix manipulations. Secondly, using high-performance computing, we develop a parallel implementation of maximum likelihood estimation for the generalised Pareto distribution. Together, these innovations allow estimation of rigorous whole-basin extreme value models in reasonable time. We evaluate the new approach in application to marginal extreme value modelling of storm peak significant wave heights in two ocean basins, accommodating spatial and directional covariate effects. © 2015 Elsevier B.V.
KW - Extremes
KW - Fast-computation
KW - High-performance computation
KW - Non-stationary effects
KW - Splines
KW - Computational efficiency
KW - Design
KW - Gaussian noise (electronic)
KW - Maximum likelihood
KW - Maximum likelihood estimation
KW - Ocean currents
KW - Offshore structures
KW - Pareto principle
KW - Storms
KW - Tensors
KW - Value engineering
KW - Fast computation
KW - Generalised Pareto distributions
KW - High performance computation
KW - High performance computing
KW - Meteorological and oceanographic data
KW - Parallel implementations
KW - Structural design
U2 - 10.1016/j.csda.2015.09.006
DO - 10.1016/j.csda.2015.09.006
M3 - Journal article
VL - 95
SP - 243
EP - 258
JO - Computational Statistics and Data Analysis
JF - Computational Statistics and Data Analysis
SN - 0167-9473
ER -