Bayesian Spatial Modeling of Extreme Precipitation Return Levels

Cooley, Daniel; Nychka, Douglas; Naveau, Philippe

doi:10.1198/016214506000000780

Cited by 466 publications

(460 citation statements)

References 2 publications

Supporting

Mentioning

451

Contrasting

Unclassified

Order By: Relevance

“…Analysis of the empirical quantiles at the stations used here confirmed this behavior. For precipitation, the form of the tail appears to vary with duration (Buishand 1991;Pearson and Henderson 1998), geographical location (Revfeim 1982;Buishand and Demaré 1990;Pearson and Henderson 1998;Friederichs 2010;Toreti et al 2010;Maraun et al 2011), and altitude (Pearson and Henderson 1998;Cooley et al 2007;Gardes and Girard 2010). For daily precipitation, the shape parameter appears to be mostly light or heavy-tailed.…”

Section: Imt Selection and Gpd Estimatesmentioning

confidence: 99%

Automatic threshold and run parameter selection: a climatology for extreme hourly precipitation in Switzerland

Fukutome

Liniger

Süveges

2014

Theor Appl Climatol

View full text Add to dashboard Cite

Extreme value analyses of a large number of relatively short time series are in increasing demand in environmental sciences and design. Here, we present an automated procedure for the peaks-over-threshold (POT) approach to extreme value theory and use it to provide a climatology of extreme hourly precipitation in Switzerland. The POT approach fits the generalized Pareto distribution (GPD) to independent exceedances above some high threshold. To guarantee independence, the time series is pruned: exceedances separated by less than a fixed interval called the run parameter are considered a cluster, and all but the cluster maxima are discarded. We propose the automation of an existing graphical method for joint selection of threshold and run parameter. Hourly precipitation is analyzed at 59 stations of the MeteoSwiss observational network over the period 1981-2010. The four seasons are considered separately. When necessary, a simple detrending is applied. Results suggest that unnecessarily large run parameters have adverse effects on the estimation of the GPD parameters. The proposed method yields mean cluster sizes that reflect the seasonal and geographical variation of lag dependence of hourly precipitation. The climatology, as represented by the return level maps and Alpine cross-section, mirror known aspects of the Swiss climate. Unlike for daily precipitation, summer thunderstorm tracks are visible in the seasonal frequency of events, rather than in the amplitude of rare events.

show abstract

Section: Imt Selection and Gpd Estimatesmentioning

confidence: 99%

Automatic threshold and run parameter selection: a climatology for extreme hourly precipitation in Switzerland

Fukutome

Liniger

Süveges

2014

Theor Appl Climatol

View full text Add to dashboard Cite

show abstract

“…A closely related approach, typically used in Bayesian modeling using Markov chain Monte Carlo simulation, is to represent variation in extremal parameters through spatial Gaussian processes. For example, Cooley et al (2007) fit rainfall exceedances using a GPD model (Eq. (3)) in which log σ follows such a process, but the shape parameter has two values, and many similar models have appeared in the literature (Coles and Casson 1998;Casson and Coles 1999;Gelfand 2009, 2010).…”

Section: Basic Ideasmentioning

confidence: 99%

“…Chavez-Demoulin et al (2011) point out that inclusion of covariates in the usual parameterization of the GPD model leads to a lack of invariance to the choice of threshold that can be resolved using the GEV parameterization. One major difference between models such as that of Cooley et al (2007) and those proposed below is that the former fit univariate extremal distributions to data at particular spatial locations, but treat the margins as independent, conditional on the covariates, so risk estimates for spatial quantities may be poor, though those at individual locations can be expected to improve because of borrowing of strength across the spatial domain. By contrast, the models described here aim to capture joint spatial properties of extremes in addition to their marginal variation.…”

Section: Basic Ideasmentioning

confidence: 99%

Geostatistics of Dependent and Asymptotically Independent Extremes

2013

View full text Add to dashboard Cite

Spatial modeling of rare events has obvious applications in the environmental sciences and is crucial when assessing the effects of catastrophic events (such as heatwaves or widespread flooding) on food security and on the sustainability of societal infrastructure. Although classical geostatistics is largely based on Gaussian processes and distributions, these are not appropriate for extremes, for which maxstable and related processes provide more suitable models. This paper provides a brief overview of current work on the statistics of spatial extremes, with an emphasis on the consequences of the assumption of max-stability. Applications to winter minimum temperatures and daily rainfall are described.

show abstract

“…Cooley et al (2007) ran simulations to show that using thresholds in the middle of the discretization interval provides numerically stable estimations, which is the approach we took here. Another possibility would be to artificially add noise to the observations to break the ties that cause the numerical difficulties (see Einmahl & Magnus 2008).…”

Section: Point Process Model Fitmentioning

confidence: 99%

Statistical modeling of hot spells and heat waves

Furrer¹,

Katz²,

Walter³

et al. 2010

Clim. Res.

View full text Add to dashboard Cite

Although hot spells and heat waves are considered extreme meteorological phenomena, the statistical theory of extreme values has only rarely, if ever, been applied. To address this shortcoming, we extended the point process approach to extreme value analysis to model the frequency, duration, and intensity of hot spells. The annual frequency of hot spells was modeled by a Poisson distribution, and their length by a geometric distribution. To account for the temporal dependence of daily maximum temperatures within a hot spell, the excesses over a high threshold were modeled by a conditional generalized Pareto distribution, whose scale parameter depends on the excess on the previous day. Requiring only univariate extreme value theory, our proposed approach is simple enough to be readily generalized to incorporate trends in hot spell characteristics. Through a heat wave simulator, the statistical modeling of hot spells can be extended to apply to more fullfledged heat waves, which are difficult to model directly. Our statistical model for hot spells was fitted to time series of daily maximum temperature during the summer heat wave season in Phoenix, Arizona (USA), Fort Collins, Colorado (USA), and Paris, France. Trends in the frequency, duration, and intensity of hot spells were fitted as well. The heat wave simulator was used to convert any such trends into the corresponding changes in the characteristics of heat waves. By being based at least in part on extreme value theory, our proposed approach is both more realistic and more flexible than techniques heretofore applied to model hot spells and heat waves.

show abstract

Bayesian Spatial Modeling of Extreme Precipitation Return Levels

Cited by 466 publications

References 2 publications

Automatic threshold and run parameter selection: a climatology for extreme hourly precipitation in Switzerland

Automatic threshold and run parameter selection: a climatology for extreme hourly precipitation in Switzerland

Geostatistics of Dependent and Asymptotically Independent Extremes

Statistical modeling of hot spells and heat waves

Contact Info

Product

Resources

About