Max-stable Models Research Articles

Assessment of alternative methods for analysing maximum rainfall spatial data based on generalized extreme value distribution

The present study aimed to analyze and spatially model maximum rainfall in the southern and southwestern regions of Minas Gerais using spatial statistical methods. Daily data on maximum rainfall were collected from 29 cities in the region. To obtain predictions of maximum rainfall for return periods of 2, 5, 10, 50, and 100 years, Bayesian Inference was employed, utilizing the most appropriate prior for each locality. The spatial analysis of the phenomenon based on results obtained through Bayesian Inference was conducted using interpolation methods, including Inverse Distance Weighting (IDW) and Kriging (Ordinary Kriging (OK) and Log-Normal Kriging (LK)). Different semivariogram models were used, and the most suitable one was selected based on cross-validation results for each method, which were also compared to those of IDW. Additionally, a spatial analysis was carried out using max-stable processes and spatial Generalized Extreme Value (GEV) distribution, with the models evaluated based on Takeuchi’s Information Criteria. All models were also assessed by calculating the mean prediction error for six locations that were not used in model fitting. The results indicated that the most suitable models among Kriging and IDW for return periods of 2, 5, and 10 years were Gaussian (LK), Spherical (OK), and Wave (OK), respectively. Among the max-stable models and spatial GEV, the most suitable for modeling was the Smith max-stable model. Consequently, for spatial prediction over 50- and 100-year return periods, OK (Wave) and the Smith max-stable model were employed.

Vecchia Likelihood Approximation for Accurate and Fast Inference with Intractable Spatial Max-Stable Models

Max-stable processes are the most popular models for high-impact spatial extreme events, as they arise as the only possible limits of spatially-indexed block maxima. However, likelihood inference for such models suffers severely from the curse of dimensionality, since the likelihood function involves a combinatorially exploding number of terms. In this article, we propose using the Vecchia approximation, which conveniently decomposes the full joint density into a linear number of low-dimensional conditional density terms based on well-chosen conditioning sets designed to improve and accelerate inference in high dimensions. Theoretical asymptotic relative efficiencies in the Gaussian setting and simulation experiments in the max-stable setting show significant efficiency gains and computational savings using the Vecchia likelihood approximation method compared to traditional composite likelihoods. Our application to extreme sea surface temperature data at more than a thousand sites across the entire Red Sea further demonstrates the superiority of the Vecchia likelihood approximation for fitting complex models with intractable likelihoods, delivering significantly better results than traditional composite likelihoods, and accurately capturing the extremal dependence structure at lower computational cost. Supplementary materials for this article are available online.

Neural networks for parameter estimation in intractable models

The goal is to use deep learning models to estimate parameters in statistical models when standard likelihood estimation methods are computationally infeasible. For instance, inference for max-stable processes is exceptionally challenging even with small datasets, but simulation is straightforward. Data from model simulations are used to train deep neural networks and learn statistical parameters from max-stable models. The proposed neural network-based method provides a competitive alternative to current approaches, as demonstrated by considerable accuracy and computational time improvements. It serves as a proof of concept for deep learning in statistical parameter estimation and can be extended to other estimation problems.

Modelling Drought Risk Using Bivariate Spatial Extremes: Application to the Limpopo Lowveld Region of South Africa

Weather and climate extremes such as heat waves, droughts and floods are projected to become more frequent and intense in several regions. There is compelling evidence indicating that changes in climate and its extremes over time influence the living conditions of society and the surrounding environment across the globe. This study applies max-stable models to capture the spatio–temporal extremes with dependence. The objective was to analyse the risk of drought caused by extremely high temperatures and deficient rainfall. Hopkin’s statistic was used to assess the clustering tendency before using the agglomerative method of hierarchical clustering to cluster the study area into n=3 temperature clusters and n=3 precipitation clusters. For the precipitation and temperature data, the values of Hopkin’s statistic were 0.7317 and 0.8446, respectively, which shows that both are significantly clusterable. Various max-stable process models were then fitted to each cluster of each variable, and the Schlather model with several covariance functions was found to be a good fit on both datasets compared to the Smith model with the Gaussian covariance function. The modelling approach presented in this paper could be useful to hydrologists, meteorologists and climatologists, including decision-makers in the agricultural sector, in enhancing their understanding of the behaviour of drought caused by extremely high temperatures and low rainfall. The modelling of these compound extremes could also assist in assessing the impact of climate change. It can be seen from this study that the size, including the topography of the location (cluster/region), provides important information about the strength of the extremal dependence.

Consistency of Bayesian inference for multivariate max-stable distributions

Predicting extreme events is important in many applications in risk analysis. Extreme-value theory suggests modelling extremes by max-stable distributions. The Bayesian approach provides a natural framework for statistical prediction. Although various Bayesian inferential procedures have been proposed in the literature of univariate extremes and some for multivariate extremes, the study of their asymptotic properties has been left largely untouched. In this paper we focus on a semiparametric Bayesian method for estimating max-stable distributions in arbitrary dimension. We establish consistency of the pertaining posterior distributions for fairly general, well-specified max-stable models, whose margins can be short-, light- or heavy-tailed. We then extend our consistency results to the case where data are samples of block maxima whose distribution is only approximately a max-stable one, which represents the most realistic inferential setting.

Spatio-temporal variability of dry and wet spells and their influence on crop yields

Spatial modelling of extreme dry (wet) spells is relevant to design agricultural insurances and establish risk management strategies. We characterised dry (wet) spells during the period 1971–2010 in 133 rainfall stations and investigated their impact on crop yields across Belgium. A day was considered as a dry (wet) day when the total amount of rainfall is below (above) 0.2 mm. A two-way ANOVA demonstrated significant differences between low and high yields across 12 agricultural regions in the number of consecutive dry (wet) days, rainfall amount during the wet days and their total during the crop growing season. Generalised Extreme Value distributions were fitted to the maxima series using maximum likelihood to estimate the distribution parameters. The number of consecutive dry (wet) days and rainfall amount was considered during the cropping season between March and October. We modelled maxima observed at rainfall stations by means of max-stable processes, an extension of multivariate extreme value theory. Spatial extreme modelling was performed through pairwise likelihood inference by a madogram and enabled 20-year return levels to be simulated across regions based on geographic coordinates and altitude as covariates. Spatial return levels confirmed the exceptionality of 2016 and 2018 as extreme wet and dry years with high return periods. The 2018 drought resulted in yield reductions of 74% for maize and 42% for potato, while the 2016 rainfall induced yield losses of up to 39% for winter cereals, 45% for maize and 53% for potato. The max-stable models provided a good fit to the joint behaviour of extremes in the number of consecutive dry (wet) days during the growing season as an indication for crop stress. The method allows for spatio-temporal characterisation of generalised extreme value distributions across regions, and provides a tool for weather based insurances and risk management.

Modeling Nonstationary Extreme Dependence With Stationary Max-Stable Processes and Multidimensional Scaling

Modeling the joint distribution of extreme events at multiple locations is a challenging task with important applications. In this study, we use max-stable models to study extreme daily precipitation events in Switzerland. The nonstationarity of the spatial process at hand involves important challenges, which are often dealt with by using a stationary model in a so-called climate space, with well-chosen covariates. Here, we instead choose to warp the weather stations under study in a latent space of higher dimension using multidimensional scaling (MDS). Two methods are proposed to define target dissimilarity matrices, based respectively on extremal coefficients and on pairwise likelihoods. Results suggest that the proposed methods allow capturing complex spatial dependences of spatial extreme precipitations, enabling in turn to reliably extrapolate functionals such as extremal coefficients. Supplemental materials for this article are available online.

Bayesian Spatial and Trend Analysis on Ozone Extreme Data in South Korea: 1991–2015

Background. Extreme events like flooding, extreme temperature, and ozone depletion are happening in every corner of the world. Thus, the need to model such rare events having enormous damage has been getting priorities in most countries of the world. Methods. The dataset contains the ozone data from 29 representative air monitoring sites in South Korea collected from 1991 to 2015. Spatial generalized extreme value (GEV) using maximum likelihood estimation (MLE) and two max-stable and Bayesian kriging models are the statistical models used for analysis. Moreover, predictive performances of these statistical models are compared using measures like root-mean-squared error (RMSE), mean absolute error (MAE), relative bias (rBIAS), and relative mean separation (rMSEP) have been utilized. Results. From the time plot of ozone data, extreme ozone concentration is increasing linearly within the specified period. The return level of ozone concentration after 10, 25, 50, and 100 years have been forecasted and showed that there was an increasing trend in ozone extremes. High spatial variability of ozone extreme was observed, and those areas around the territories were having extreme ozone concentration than the centers. Moreover, Bayesian Kriging brought about relatively the minimum RMSE compared to the other models. Conclusion. The extreme ozone concentration has clearly showed a positive trend and spatial variation. Moreover, among the models considered in the paper, the Bayesian Kriging has been chosen as the better model.

A regionalisation approach for rainfall based on extremal dependence

To mitigate the risk posed by extreme rainfall events, we require statistical models that reliably capture extremes in continuous space with dependence. However, assuming a stationary dependence structure in such models is often erroneous, particularly over large geographical domains. Furthermore, there are limitations on the ability to fit existing models, such as max-stable processes, to a large number of locations. To address these modelling challenges, we present a regionalisation method that partitions stations into regions of similar extremal dependence using clustering. To demonstrate our regionalisation approach, we consider a study region of Australia and discuss the results with respect to known climate and topographic features. To visualise and evaluate the effectiveness of the partitioning, we fit max-stable models to each of the regions. This work serves as a prelude to how one might consider undertaking a project where spatial dependence is non-stationary and is modelled on a large geographical scale.

Composite likelihood methods for histogram-valued random variables

Symbolic data analysis has been proposed as a technique for summarising large and complex datasets into a much smaller and tractable number of distributions—such as random rectangles or histograms—each describing a portion of the larger dataset. Recent work has developed likelihood-based methods that permit fitting models for the underlying data while only observing the distributional summaries. However, while powerful, when working with random histograms this approach rapidly becomes computationally intractable as the dimension of the underlying data increases. We introduce a composite-likelihood variation of this likelihood-based approach for the analysis of random histograms in K dimensions, through the construction of lower-dimensional marginal histograms. The performance of this approach is examined through simulated and real data analysis of max-stable models for spatial extremes using millions of observed datapoints in more than $$K=100$$ dimensions. Large computational savings are available compared to existing model fitting approaches.

Projecting Flood-Inducing Precipitation with a Bayesian Analogue Model

The hazard of pluvial flooding is largely influenced by the spatial and temporal dependence characteristics of precipitation. When extreme precipitation possesses strong spatial dependence, the risk of flooding is amplified due to catchment factors such as topography that cause runoff accumulation. Temporal dependence can also increase flood risk as storm water drainage systems operating at capacity can be overwhelmed by heavy precipitation occurring over multiple days. While transformed Gaussian processes are common choices for modeling precipitation, their weak tail dependence may lead to underestimation of flood risk. Extreme value models such as the generalized Pareto processes for threshold exceedances and max-stable models are attractive alternatives, but are difficult to fit when the number of observation sites is large, and are of little use for modeling the bulk of the distribution, which may also be of interest to water management planners. While the atmospheric dynamics governing precipitation are complex and difficult to fully incorporate into a parsimonious statistical model, non-mechanistic analogue methods that approximate those dynamics have proven to be promising approaches to capturing the temporal dependence of precipitation. In this paper, we present a Bayesian analogue method that leverages large, synoptic-scale atmospheric patterns to make precipitation forecasts. Changing spatial dependence across varying intensities is modeled as a mixture of spatial Student-t processes that can accommodate both strong and weak tail dependence. The proposed model demonstrates improved performance at capturing the distribution of extreme precipitation over Community Atmosphere Model (CAM) 5.2 forecasts.Supplementary materials accompanying this paper appear online.

Distributionally robust inference for extreme Value-at-Risk

Under general multivariate regular variation conditions, the extreme Value-at-Risk of a portfolio can be expressed as an integral of a known kernel with respect to a generally unknown spectral measure supported on the unit simplex. The estimation of the spectral measure is challenging in practice and virtually impossible in high dimensions. This motivates the problem studied in this work, which is to find universal lower and upper bounds of the extreme Value-at-Risk under practically estimable constraints. That is, we study the infimum and supremum of the extreme Value-at-Risk functional, over the infinite dimensional space of all possible spectral measures that meet a finite set of constraints. We focus on extremal coefficient constraints, which are popular and easy to interpret in practice. Our contributions are twofold. First, we show that optimization problems over an infinite dimensional space of spectral measures are in fact dual problems to linear semi-infinite programs (LSIPs) – linear optimization problems in Euclidean space with an uncountable set of linear constraints. This allows us to prove that the optimal solutions are in fact attained by discrete spectral measures supported on finitely many atoms. Second, in the case of balanced portfolia, we establish further structural results for the lower bounds as well as closed form solutions for both the lower- and upper-bounds of extreme Value-at-Risk in the special case of a single extremal coefficient constraint. The solutions unveil important connections to the Tawn–Molchanov max-stable models. The results are illustrated with two applications: a real data example and closed-form formulae in a market plus sectors framework.

Strong convergence of multivariate maxima

AbstractIt is well known and readily seen that the maximum of n independent and uniformly on [0, 1] distributed random variables, suitably standardised, converges in total variation distance, as n increases, to the standard negative exponential distribution. We extend this result to higher dimensions by considering copulas. We show that the strong convergence result holds for copulas that are in a differential neighbourhood of a multivariate generalised Pareto copula. Sklar’s theorem then implies convergence in variational distance of the maximum of n independent and identically distributed random vectors with arbitrary common distribution function and (under conditions on the marginals) of its appropriately normalised version. We illustrate how these convergence results can be exploited to establish the almost-sure consistency of some estimation procedures for max-stable models, using sample maxima.

Editorial of special issue on statistics for climate change and the environment

Editorial of special issue on statistics for climate change and the environment

Modeling the spatial dependence of floods using the Fisher copula

Abstract. Floods often affect not only a single location, but also a whole region. Flood frequency analysis should therefore be undertaken at a regional scale which requires the considerations of the dependence of events at different locations. This dependence is often neglected even though its consideration is essential to derive reliable flood estimates. A model used in regional multivariate frequency analysis should ideally consider the dependence of events at multiple sites which might show dependence in the lower and/or upper tail of the distribution. We here seek to propose a simple model that on the one hand considers this dependence with respect to the network structure of the region and on the other hand allows for the simulation of stochastic event sets at both gauged and ungauged locations. The new Fisher copula model is used for representing the spatial dependence of flood events in the nested Thur catchment in Switzerland. Flood event samples generated for the gauged stations using the Fisher copula are compared to samples generated by other dependence models allowing for modeling of multivariate data including elliptical copulas, R-vine copulas, and max-stable models. The comparison of the dependence structures of the generated samples shows that the Fisher copula is a suitable model for capturing the spatial dependence in the data. We therefore use the copula in a way such that it can be used in an interpolation context to simulate event sets comprising gauged and ungauged locations. The spatial event sets generated using the Fisher copula well capture the general dependence structure in the data and the upper tail dependence, which is of particular interest when looking at extreme flood events and when extrapolating to higher return periods. The Fisher copula was for a medium-sized catchment found to be a suitable model for the stochastic simulation of flood event sets at multiple gauged and ungauged locations.

Modeling snow depth extremes in Austria

Maps of extreme snow depths are important for structural design and general risk assessment in mountainous countries like Austria. The smooth modeling approach is commonly accepted to provide more accurate margins than max-stable processes. In contrast, max-stable models allow for risk estimation due to explicitly available spatial extremal dependencies, in particular when anisotropy is accounted for. However, the difference in return levels is unclear, when modeled smoothly or with max-stable processes. The objective of this study is twofold: first, to investigate that question and to provide snow depth return level maps for Austria; and second, to investigate spatial dependencies of extreme snow depths in Austria in detail and to find a suitable model for risk estimation. Therefore, a model selection procedure was used to define a marginal model for the GEV parameters. This model was fitted to 210 snow depth series comprising a length of 42 years using the smooth model approach and different max-stable models allowing for anisotropy. Despite relatively clear advantages for the Extremal-t max-stable process based on two scores compared to the smooth model as well as the Brown–Resnick, Geometric Gaussian and Schlather processes, the difference in 100-year snow depth return levels is too small, to decide which approach works better. Spatial dependencies of snow depth extremes between the regions north and south of the Austrian Alps are almost independent. Dependencies are stronger in the south for small distances between station pairs up to 120 km and become stronger in the north for larger distances. For risk modeling the Austrian Alps could be separated into regions north and south of the Alps. Fitting an anisotropic Extremal-t max-stable process to either side of the Alps can improve modeling of joint exceedance probabilities compared to one single model for the whole of Austria, especially for small station distances.

Geometric ergodicity for some space–time max-stable Markov chains

Max-stable processes are central models for spatial extremes. In this paper, we focus on some space–time max-stable models introduced in Embrechts et al. (2016). The processes considered induce discrete-time Markov chains taking values in the space of continuous functions from the unit sphere of R3 to (0,∞). We show that these Markov chains are geometrically ergodic. An interesting feature lies in the fact that the state space is not locally compact, making the classical methodology inapplicable. Instead, we use the fact that the state space is Polish and apply results presented in Hairer (2010).

ABC model selection for spatial extremes models applied to South Australian maximum temperature data

Max-stable processes are a common choice for modelling spatial extreme data as they arise naturally as the infinite-dimensional generalisation of multivariate extreme value theory. Statistical inference for such models is complicated by the intractability of the multivariate density function. Nonparametric, composite likelihood-based, and Bayesian approaches have been proposed to address this difficulty. More recently, a simulation-based approach using approximate Bayesian computation (ABC) has been employed for estimating parameters of max-stable models. ABC algorithms rely on the evaluation of discrepancies between model simulations and the observed data rather than explicit evaluations of computationally expensive or intractable likelihood functions. The use of an ABC method to perform model selection for max-stable models is explored. Three max-stable models are regarded: the extremal-t model with either a Whittle–Matérn or a powered exponential covariance function, and the Brown–Resnick model with power variogram. In addition, the non-extremal Student-t copula model with a Whittle–Matérn or a powered exponential covariance function is also considered. The method is applied to annual maximum temperature data from 25 weather stations dispersed around South Australia.

Probabilities of Concurrent Extremes

ABSTRACTThe statistical modeling of spatial extremes has been an active area of recent research with a growing domain of applications. Much of the existing methodology, however, focuses on the magnitudes of extreme events rather than on their timing. To address this gap, this article investigates the notion of extremal concurrence. Suppose that daily temperatures are measured at several synoptic stations. We say that extremes are concurrent if record maximum temperatures occur simultaneously, that is, on the same day for all stations. It is important to be able to understand, quantify, and model extremal concurrence. Under general conditions, we show that the finite sample concurrence probability converges to an asymptotic quantity, deemed extremal concurrence probability. Using Palm calculus, we establish general expressions for the extremal concurrence probability through the max-stable process emerging in the limit of the component-wise maxima of the sample. Explicit forms of the extremal concurrence probabilities are obtained for various max-stable models and several estimators are introduced. In particular, we prove that the pairwise extremal concurrence probability for max-stable vectors is precisely equal to the Kendall’s τ. The estimators are evaluated from simulations and applied to study temperature extremes in the United States. Results demonstrate that concurrence probability can be used to study, for example, the effect of global climate phenomena such as the El Niño Southern Oscillation (ENSO) or global warming on the spatial structure and areal impact of extremes.

Extremal attractors of Liouville copulas

Liouville copulas, which were introduced in McNeil and Neslehova (2010), are asymmetric generalizations of the ubiquitous Archimedean copula class. They are the dependence structures of scale mixtures of Dirichlet distributions, also called Liouville distributions. In this paper, the limiting extreme-value copulas of Liouville copulas and of their survival counterparts are derived. The limiting max-stable models, termed here the scaled extremal Dirichlet, are new and encompass several existing classes of multivariate max-stable distributions, including the logistic, negative logistic and extremal Dirichlet. As shown herein, the stable tail dependence function and angular density of the scaled extremal Dirichlet model have a tractable form, which in turn leads to a simple de Haan representation. The latter is used to design efficient algorithms for unconditional simulation based on the work of Dombry, Engelke and Oesting (2015) and to derive tractable formulas for maximum-likelihood inference. The scaled extremal Dirichlet model is illustrated on river flow data of the river Isar in southern Germany.