Schedule for: 16w5092 - Uncertainty Modeling in the Analysis of Weather, Climate and Hydrological Extremes

A buffet dinner is served daily between 5:30pm and 7:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Rainfall extremes appear to be changing around the world but there is little information on how extreme short-duration events might change. This talk will present results from the NERC-funded CONVEX project which produced the first climate model integrations at convective-permitting scales. The 1.5km regional climate model run over the southern half of the UK showed that extreme rainfall events at the hourly scale are expected to increase in frequency in the future during summer months. Additional analysis of the model, and other convective-permitting model runs, show that these models can reproduce observed relationships between temperature and precipitation extremes and extreme value scaling. Discussion will also be made of temperature-precipitation scaling relationships from ad-hoc sub-daily observational studies and how these vary globally and how climate models suggest these should change in future from the few studies that have now been performed across the globe. The first results from the ERC-funded INTENSE project, which aims to perform a global analysis of changes to sub-daily precipitation extremes, will be presented, as well as progress in this area in the international arena through the GEWEX sub-daily precipitation cross-cut activity.

Agata Imielska, Communicating Extremes: Realising the value of climate information \[ \] Abstract: When it starts pouring rain, coming down with hail, or a drought sets in, the community and media turn to experts to understand what is happening to their environment. It is important to be ready to effectively engage with our community providing information that is pertinent to the issues at hand, but in a way that will be easily understood. The impact of extreme events opens the communication channels which may otherwise be closed. However, misinterpreting good information can be worse than not having the information at all. So how do we realise the full benefit of climate information? Multichannel communication is critically important, as is utilising different communication mediums such as websites, blogs, social media, video as well as reports and face-to-face presentations. Information needs to be tailored to users needs to ensure the information is well understood and correctly integrated into decision making frameworks and for the full value of the information to be realised. \[-----------\] Michael Scheuerer (with T.M. Hamill and R.S. Webb), Using GEFS ensemble forecasts for decision making in reservoir management in California \[ \] Abstract: Reservoirs such as Lake Mendocino in California's Russian River Basin provide flood control, water supply, recreation, and environmental stream flow regulation. Many of these reservoirs are operated by the U.S. Army Corps of Engineers (Corps) according to water control manuals that specify elevations for an upper volume of reservoir storage that must be kept available for capturing storm runoff and reducing flood risk, and a lower volume of storage that may be used for water supply. During extreme rainfall events, runoff is captured by these reservoirs and released as quickly as possible to create flood storage space for another potential storm. These flood control manuals are based on typical historical weather patterns - wet during the winter, dry otherwise - but are not informed directly by weather prediction. Alternative reservoir management approaches such as Forecast-Informed Reservoir Operations (FIRO), which seek to incorporate advances in weather prediction, are currently being explored as means to improve water supply availability while maintaining flood risk reduction and providing additional ecosystem benefits. \[ \] We present results from a FIRO proof-of-concept study investigating the reliability of post-processed GEFS ensemble forecasts to predict the probability that day 6-to-10 precipitation accumulations in certain areas in California exceed a high threshold. Our results suggest that reliable forecast guidance can be provided, and the resulting probabilities could be used to inform decisions to release or hold water in the reservoirs. We illustrate the potential of these forecasts in a case study of extreme event probabilities for the Russian River Basin in California. \[----------\] Xuebin Zhang (with M. Shephard, G. Li, J. Cole, Y. Jiao, J. Scinocca), Probable Maximum Precipitation Response to Projected Climate Change over North America

A buffet dinner is served daily between 5:30pm and 7:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

Detection and attribution of climate change has been a growing activity since the 90's when the question of a possible human influence on the observed climate arose. I will first briefly introduce this theme together with the standard definitions of detection and attribution. Second, I will review the statistical models that have been used over the last 20 years to deal with these questions. Those models were predominantly linear regression models where the observations are regressed onto expected response patterns to different external forcings. Several levels of complexity have been proposed, from usual linear regression models to sophisticated error-in-variable models where observational and climate modelling uncertainties are accounted for. A recent, simple alternative proposes to avoid using linear regression based on similar assumption regarding uncertainties. Third, I will discuss a few statistical issues common to those models. A first issue involves the estimation of internal variability and its covariance matrix, which is required to carry out optimal inference. A second issue involves the estimation of climate modelling uncertainty, and the underlying assumptions. A third issue involves the data preprocessing and the dimension reduction that is needed to use those models.

Meet in foyer of TCPL to participate in the BIRS group photo. The photograph will be taken outdoors, so dress appropriately for the weather. Please don't be late, or you might not be in the official group photo!

Julien Cattiaux: Estimating the contribution of atmospheric dynamics to extreme events \[----------\] Debbie Polson: Impacts of local and remote anthropogenic aerosols on the 20th century West Africa and South Asia monsoons \[----------\] Michael Wehner: Extreme event attribution statements for very rare events

The last few years have seen very rapid advances in modelling complex extreme events, such as extreme spatial rainfall, wind- storms and the like. Much of this advance has been based on the use of max-stable processes for modelling maxima, but more recent work has concerned peaks over thresholds modelling, which is known to be more e cient in simpler settings. It turns out that it is both more e cient and somewhat easier in complex settings also. This talk will survey recent advances in this area, with special reference to modelling extreme rainfall.

We will explore the link between rare recurrences of states in dynamical systems and the extreme value theory. Statistics is used to infer important properties of the system such as the distribution of local dimensions (entropy) and the stability of all accessible states. We apply this framework to weather reanalysis data, the daily NCEP 1948-2015 sea level pressure (SLP) fields over the North Atlantic. We show that extremes in the local (daily) dimensions and in the stability of SLP fields are related to climate extremes such as historical storms and blocking.

Mohamed Ali Ben Alaya, Non-Gaussian multisite simulation of extreme daily precipitation: downscaling application (with T.B.M.J. Ouarda and F. Chebana) \[ \] Abstract: Probabilistic regression approaches for downscaling daily precipitation are very useful. They provide the entire conditional distribution at each forecast step which ensures the preservation of the temporal variability. The question addressed in this study is: How to simulate multisite characteristics of extreme daily precipitation from probabilistic regression models? Recent publications point out to the complexity of multisite properties of extreme daily precipitation and highlight the need of using non-Gaussian flexible tools. This work proposes a fair compromise between simplicity and flexibility avoiding model misspecification. A suitable nonparametric bootstrapping (NB) technique is adopted. A downscaling model which merges a vector generalized linear model (VGLM as a probabilistic regression tool using the Generalized-Pareto distribution) and the proposed bootstrapping technique is introduced to simulate realistic multisite precipitation series. The model is applied to data sets from the southern part of the province of Quebec, Canada. It is shown that the model is capable of reproducing both at-site and multisite properties of extreme daily precipitation. Results indicate the superiority of the proposed NB technique, over a multivariate autoregressive Gaussian framework (i.e. Gaussian copula). \[----------\] Alex Cannon, Temporal scaling of sub-daily precipitation extremes in climate models: historical evaluation and future projections \[----------\] Robert Yuen, On the aggregation of dependent catastrophic risks \[ \] Abstract: We establish the solution to semi-infinate optimization programs that determine exact bounds on the extreme value-at-risk for the sum of regularly varying dependent losses under partial dependence assumptions. Such problems are important in determining minimum solvency requirements for catastrophic perils. We show that the theoretical range of value-at-risk is significantly reduced when making a small number of partial dependence assumptions on a portfolio of risks.

Meet in the Corbett Hall Lounge for a guided tour of The Banff Centre campus.

Spatial statistics plays an important role in weather and climate modelling, especially as smoothing or prediction techniques for generating gridded datasets or exploring stochastically simulated what-if scenarios. Important in all goals is uncertainty quantification in terms of probabilistic statements. Modern spatial statistics has shifted focus to nonstationary modelling, methods for large datasets and multivariate processes. This talk will cover a brief introduction to each of these topics with the goal of generating some intuition.

Max-stable distributions are widely used for the modeling of multivariate extreme events, as they arise as natural limits of renormalized componentwise maxima of random vectors. However, when the dimension is large, the number of terms involved in the likelihood function becomes extremely large, making it intractable for classical inference. In practice, composite likelihoods are often used instead, but suffer from a loss in efficiency. In this talk, an alternative approach to perform full likelihood inference based on an EM algorithm is explored, where an additional random partition associated to the occurrence times of maxima is introduced. Treating this partition as a missing observation, the completed likelihood becomes simple and a (stochastic) EM algorithm may be used to maximize the full likelihood. The performance of this novel approach will be illustrated with numerical results based on the logistic model. \[ \[ Joint work with Clement Dombry, Marc Genton and Mathieu Ribatet. \[ \] Reference : Ailliot P., Delyon B., Monbet V., Prevosto M. Dependent time changed processes with applications to nonlinear ocean waves. arXiv:1510.02302,

Sebastian Engelke, Statistical regionalization for estimation of extreme river discharges (with Peiman Asadi) \[ \] Abstract: The accurate quantification of peak flow values with long return periods is crucial for national agencies in order to design effective flood protection and reduce economic and ecological costs. For gauging stations on a river network with long discharge records univariate extreme value theory provides reliable tools for model fitting, identification and assessment of parameter uncertainty. At ungauged locations, however, where no observations are available, these methods are no longer applicable. We propose a statistical regionalization approach that identifies the optimal region with gauged stations that are hydrologically similar to the target ungauged location. A regression model is then used to transfer information on extreme discharges from this region of influence to the ungauged site. This estimation procedure is applied to discharge data on Swiss river networks and is compared to competing methods. \[----------\] Raphael Huser, Factor copula models for spatial data \[ \] Abstract: We propose a new copula model that can be used with spatial data. Unlike the multivariate normal copula, the proposed copula is based on the assumption that a common factor exists and affects the joint dependence of all measurements of the process. Moreover, the proposed copula can model tail dependence and tail asymmetry. The model is parameterized in terms of a covariance function that may be chosen from the many models proposed in the literature, such as the Matern model. For some choice of common factors, the joint copula density is given in closed form and therefore likelihood estimation is very fast. In the general case, one-dimensional numerical integration is needed to calculate the likelihood, but estimation is still reasonably fast even with large data sets. We use simulation studies to show the wide range of dependence structures that can be generated by the proposed model with different choices of common factors. We apply the proposed model to spatial temperature data and compare its performance with some popular geostatistics models. \[----------\] Bjoern Kriesche, A model-based approach to the computation of area probabilities for precipitation exceeding a certain threshold \[----------\] Emeric Thibaud, A space-time max-stable model for extreme low temperatures in northern Finland \[ \] Abstract: Models for spatial extremes must account appropriately for asymptotic dependence, and this motivates the use of max-stable processes, which are the only non-trivial limits of properly rescaled pointwise maxima of random functions. In this poster I describe the construction of a Bayesian hierarchical model for extreme low temperatures in northern Finland, in order to consider the likely effects of climate change on the risk of forest damage caused by outbreaks of pest insect populations. \[----------\] Phyllis Wan (with R. Davis), Applications of distance covariance in extreme value analysis

The extremes community has developed elegant theory for spatial processes and more recently computational methods to handle small- to moderately-sized datasets. In this talk we consider two fully-Bayesian approaches to analyze large datasets covering thousands of spatial locations often seen in modern applications. In the first approach we use data-driven basis functions and a low-rank max-stable process to efficiently represent spatial variability. In the second approach we used skewed-t processes which capture extremal dependence with computation on the order of a Gaussian analysis. These methods are used to study extreme precipitation, air pollution, and forest fire events. This is joint work with Sam Morris (NCSU), Dan Cooley (CSU), and Emeric Thibaud (CSU).