Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Posts

publications

Causal Modelling of Heavy-Tailed Variables and Confounders with Application to River Flow

Published in Extremes, 2023

Confounding variables are a recurrent challenge for causal discovery and inference. In many situations, complex causal mechanisms only manifest themselves in extreme events, or take simpler forms in the extremes. Stimulated by data on extreme river flows and precipitation, we introduce a new causal discovery methodology for heavy-tailed variables that allows the effect of a known potential confounder to be almost entirely removed when the variables have comparable tails, and also decreases it sufficiently to enable correct causal inference when the confounder has a heavier tail. We also introduce a new parametric estimator for the existing causal tail coefficient and a permutation test. Simulations show that the methods work well and the ideas are applied to the motivating dataset.

Recommended citation: Pasche, O. C., Chavez-Demoulin, V. and Davison, A. C. (2023). "Causal modelling of heavy-tailed variables and confounders with application to river flow." Extremes 26(3), 573–594. https://doi.org/10.1007/s10687-022-00456-4
Download Paper

The Effect of a Short Observational Record on the Statistics of Temperature Extremes

Published in Geophysical Research Letters, 2023

In June 2021, the Pacific Northwest experienced a heatwave that broke all previous records. Estimated return levels based on observations up to the year before the event suggested that reaching such high temperatures is not possible in today’s climate. We here assess the suitability of the prevalent statistical approach by analyzing extreme temperature events in climate model large ensemble and synthetic extreme value data. We demonstrate that the method is subject to biases, as high return levels are generally underestimated and, correspondingly, the return period of low-likelihood heatwave events is overestimated, if the underlying extreme value distribution is derived from a short historical record. These biases have even increased in recent decades due to the emergence of a pronounced climate change signal. Furthermore, if the analysis is triggered by an extreme event, the implicit selection bias affects the likelihood assessment depending on whether the event is included in the modeling.

Recommended citation: Zeder, J., Sippel, S., Pasche, O. C., Engelke, S., and Fischer, E. M. (2023). "The effect of a short observational record on the statistics of temperature extremes." Geophysical Research Letters 50(16), e2023GL104090. https://doi.org/10.1029/2023GL104090
Download Paper

Validating Deep-Learning Weather Forecast Models on Recent High-Impact Extreme Events

Published in Artificial Intelligence for the Earth Systems, 2024

The forecast accuracy of machine learning (ML) weather prediction models is improving rapidly, leading many to speak of a “second revolution in weather forecasting”. With numerous methods being developed, and limited physical guarantees offered by ML models, there is a critical need for comprehensive evaluation of these emerging techniques. While this need has been partly fulfilled by benchmark datasets, they provide little information on rare and impactful extreme events, or on compound impact metrics, for which model accuracy might degrade due to misrepresented dependencies between variables. To address these issues, we compare ML weather prediction models (GraphCast, PanguWeather, FourCastNet) and ECMWF’s high-resolution forecast (HRES) system in three case studies: the 2021 Pacific Northwest heatwave, the 2023 South Asian humid heatwave, and the North American winter storm in 2021. We find that ML weather prediction models locally achieve similar accuracy to HRES on the record-shattering Pacific Northwest heatwave, but under-perform when aggregated over space and time. However, they forecast the compound winter storm substantially better. We also highlight structural differences in how the errors of HRES and the ML models build up to that event. The ML forecasts lack important variables for a detailed assessment of the health risks of the 2023 humid heatwave. Using a possible substitute variable, prediction errors show spatial patterns with the highest danger levels over Bangladesh being underestimated by the ML models. Generally, case-study-driven, impact-centric evaluation can complement existing research, increase public trust, and aid in developing reliable ML weather prediction models.

Recommended citation: Pasche, O. C., Wider, J., Zhang, Z., Zscheischler, J. and Engelke, S. (2024). "Validating Deep-Learning Weather Forecast Models on Recent High-Impact Extreme Events." Artificial Intelligence for the Earth Systems. http://doi.org/10.1175/AIES-D-24-0033.1
Download Paper

Modeling Extreme Events: Univariate and Multivariate Data-Driven Approaches

Published in Extremes, 2024

This article summarizes the contribution of team genEVA to the EVA (2023) Conference Data Challenge. The challenge comprises four individual tasks, with two focused on univariate extremes and two related to multivariate extremes. In the first univariate assignment, we estimate a conditional extremal quantile using a quantile regression approach with neural networks. For the second, we develop a fine-tuning procedure for improved extremal quantile estimation with a given conservative loss function. In the first multivariate sub-challenge, we approximate the data-generating process with a copula model. In the remaining task, we use clustering to separate a high-dimensional problem into approximately independent components. Overall, competitive results were achieved for all challenges, and our approaches for the univariate tasks yielded the most accurate quantile estimates in the competition.

Recommended citation: Buriticá, G., Hentschel, M., Pasche, O. C., Röttger, F. and Zhang, Z. (2024). "Modeling extreme events: Univariate and multivariate data-driven approaches." Extremes. https://doi.org/10.1007/s10687-024-00499-9
Download Paper

Granger Causality in Extremes

Published in ArXiv Preprint, 2024

We propose a mathematical framework for Granger causality in extremes, designed to identify causal links from extreme events in time series. Granger causality plays a pivotal role in understanding directional relationships among time-varying variables. While the task of causal discovery in time series gains heightened importance during extreme and highly volatile periods, state-of-the-art methods primarily focus on causality within the body of the distribution, often overlooking causal mechanisms that manifest only during extreme events. Our framework is designed to infer causality mainly from extreme events by leveraging the causal tail coefficient. We establish equivalences between causality in extremes and other causal concepts, including (classical) Granger causality, Sims causality, and structural causality. We prove other key properties of Granger causality in extremes and show that the framework is especially helpful in the presence of hidden confounders. We also propose a novel inference method for detecting the presence of Granger causality in extremes from observational data. Our method is model-free, can handle non-linear and high-dimensional time series, outperforms current state-of-the-art methods in all considered setups, both in performance and speed, and was found to uncover coherent effects when applied to financial and extreme weather problems. An open-source implementation of our proposed methodology is provided.

Recommended citation: Bodik J., Pasche, O. C. (2024). "Granger Causality in Extremes." ArXiv:2407.09632. https://doi.org/10.48550/arXiv.2407.09632
Download Paper

Neural Networks for Extreme Quantile Regression with an Application to Forecasting of Flood Risk

Published in Annals of Applied Statistics, 2024

Risk assessment for extreme events requires accurate estimation of high quantiles that go beyond the range of historical observations. When the risk depends on the values of observed predictors, regression techniques are used to interpolate in the predictor space. We propose the EQRN model that combines tools from neural networks and extreme value theory into a method capable of extrapolation in the presence of complex predictor dependence. Neural networks can naturally incorporate additional structure in the data. We develop a recurrent version of EQRN that is able to capture complex sequential dependence in time series. We apply this method to forecast flood risk in the Swiss Aare catchment. It exploits information from multiple covariates in space and time to provide one-day-ahead predictions of return levels and exceedance probabilities. This output complements the static return level from a traditional extreme value analysis, and the predictions are able to adapt to distributional shifts as experienced in a changing climate. Our model can help authorities to manage flooding more effectively and to minimize their disastrous impacts through early warning systems.

Recommended citation: Pasche, O. C. and Engelke, S. (2024). "Neural networks for extreme quantile regression with an application to forecasting of flood risk." Annals of Applied Statistics 18(4), 2818–2839. https://doi.org/10.1214/24-AOAS1907
Download Paper

software

EQRN R Package

Extreme Quantile Regression Neural Networks for Conditional Risk Assessment

talks

Extreme Value Analysis (EVA) 2021

Published: July 02, 2021

Valpred 3

Published: April 01, 2022

CRG Workshop 2022

Published: June 01, 2022

CUSO Ph.D. Day 2022

Published: June 01, 2022

International Symposium on Nonparametric Statistics (ISNPS) 2022

Published: June 21, 2022

CMStatistics 2022

Published: December 19, 2022

CRG Workshop 2023

Published: February 01, 2023

CUSO Career Days 2023

Published: February 10, 2023

Valpred 4

Published: April 04, 2023

CUSO Ph.D. Day 2023

Published: June 01, 2023

Extreme Value Analysis (EVA) 2023

Published: June 30, 2023

Invited talk in the session “Extremes and machine learning”, organized and chaired by Olivier Wintenberger.
Invited talk in the “EVA (2023) Data Challenge” session, organized and chaired by Christian Rohrbeck, Emma Simpson and Jonathan Tawn.