Search

Accurate precipitation forecasts have a high socio-economic value due to their role in decision-making in various fields such as transport networks and farming. We propose a global statistical postprocessing method for grid-based precipitation ensemble forecasts. This U-Net-based distributional regression method predicts marginal distributions in the form of parametric distributions inferred by scoring rule minimization. Distributional regression U-Nets are compared to state-of-the-art postprocessing methods for daily 21-h forecasts of 3-h accumulated precipitation over the South of France. Training data comes from the M\'et\'eo-France weather model AROME-EPS and spans 3 years. A practical challenge appears when consistent data or reforecasts are not available. Distributional regression U-Nets compete favorably with the raw ensemble. In terms of continuous ranked probability score, they reach a performance comparable to quantile regression forests (QRF). However, they are unable to provide calibrated forecasts in areas associated with high climatological precipitation. In terms of predictive power for heavy precipitation events, they outperform both QRF and semi-parametric QRF with tail extensions., Comment: for associated code, see https://github.com/pic-romain/unet-pp

Proper scoring rules are an essential tool to assess the predictive performance of probabilistic forecasts. However, propriety alone does not ensure an informative characterization of predictive performance and it is recommended to compare forecasts using multiple scoring rules. With that in mind, interpretable scoring rules providing complementary information are necessary. We formalize a framework based on aggregation and transformation to build interpretable multivariate proper scoring rules. Aggregation-and-transformation-based scoring rules are able to target specific features of the probabilistic forecasts; which improves the characterization of the predictive performance. This framework is illustrated through examples taken from the literature and studied using numerical experiments showcasing its benefits. In particular, it is shown that it can help bridge the gap between proper scoring rules and spatial verification tools., Comment: for associated code, see https://github.com/pic-romain/aggregation-transformation

Causal asymmetry is based on the principle that an event is a cause only if its absence would not have been a cause. From there, uncovering causal effects becomes a matter of comparing a well-defined score in both directions. Motivated by studying causal effects at extreme levels of a multivariate random vector, we propose to construct a model-agnostic causal score relying solely on the assumption of the existence of a max-domain of attraction. Based on a representation of a Generalized Pareto random vector, we construct the causal score as the Wasserstein distance between the margins and a well-specified random variable. The proposed methodology is illustrated on a hydrologically simulated dataset of different characteristics of catchments in Switzerland: discharge, precipitation, and snowmelt.

Generating accurate extremes from an observational data set is crucial when seeking to estimate risks associated with the occurrence of future extremes which could be larger than those already observed. Applications range from the occurrence of natural disasters to financial crashes. Generative approaches from the machine learning community do not apply to extreme samples without careful adaptation. Besides, asymptotic results from extreme value theory (EVT) give a theoretical framework to model multivariate extreme events, especially through the notion of multivariate regular variation. Bridging these two fields, this paper details a variational autoencoder (VAE) approach for sampling multivariate heavy-tailed distributions, i.e., distributions likely to have extremes of particularly large intensities. We illustrate the relevance of our approach on a synthetic data set and on a real data set of discharge measurements along the Danube river network. The latter shows the potential of our approach for flood risks' assessment. In addition to outperforming the standard VAE for the tested data sets, we also provide a comparison with a competing EVT-based generative approach. On the tested cases, our approach improves the learning of the dependency structure between extremes.

In extreme value theory and other related risk analysis fields, probability weighted moments (PWM) have been frequently used to estimate the parameters of classical extreme value distributions. This method-of-moment technique can be applied when second moments are finite, a reasonable assumption in many environmental domains like climatological and hydrological studies. Three advantages of PWM estimators can be put forward: their simple interpretations, their rapid numerical implementation and their close connection to the well-studied class of U-statistics. Concerning the later, this connection leads to precise asymptotic properties, but non asymptotic bounds have been lacking when off-the-shelf techniques (Chernoff method) cannot be applied, as exponential moment assumptions become unrealistic in many extreme value settings. In addition, large values analysis is not immune to the undesirable effect of outliers, for example, defective readings in satellite measurements or possible anomalies in climate model runs. Recently, the treatment of outliers has sparked some interest in extreme value theory, but results about finite sample bounds in a robust extreme value theory context are yet to be found, in particular for PWMs or tail index estimators. In this work, we propose a new class of robust PWM estimators, inspired by the median-of-means framework of Devroye et al. (2016). This class of robust estimators is shown to satisfy a sub-Gaussian inequality when the assumption of finite second moments holds. Such non asymptotic bounds are also derived under the general contamination model. Our main proposition confirms theoretically a trade-off between efficiency and robustness. Our simulation study indicates that, while classical estimators of PWMs can be highly sensitive to outliers.

The statistical modeling of discrete extremes has received less attention than their continuous counterparts in the Extreme Value Theory (EVT) literature. One approach to the transition from continuous to discrete extremes is the modeling of threshold exceedances of integer random variables by the discrete version of the generalized Pareto distribution. However, the optimal choice of thresholds defining exceedances remains a problematic issue. Moreover, in a regression framework, the treatment of the majority of non-extreme data below the selected threshold is either ignored or separated from the extremes. To tackle these issues, we expand on the concept of employing a smooth transition between the bulk and the upper tail of the distribution. In the case of zero inflation, we also develop models with an additional parameter. To incorporate possible predictors, we relate the parameters to additive smoothed predictors via an appropriate link, as in the generalized additive model (GAM) framework. A penalized maximum likelihood estimation procedure is implemented. We illustrate our modeling proposal with a real dataset of avalanche activity in the French Alps. With the advantage of bypassing the threshold selection step, our results indicate that the proposed models are more flexible and robust than competing models, such as the negative binomial distribution, Comment: 19 pages, 6 figures including and 3 Tables

The theoretical advances on the properties of scoring rules over the past decades have broadened the use of scoring rules in probabilistic forecasting. In meteorological forecasting, statistical postprocessing techniques are essential to improve the forecasts made by deterministic physical models. Numerous state-of-the-art statistical postprocessing techniques are based on distributional regression evaluated with the Continuous Ranked Probability Score (CRPS). However, theoretical properties of such evaluation with the CRPS have solely considered the unconditional framework (i.e. without covariates) and infinite sample sizes. We extend these results and study the rate of convergence in terms of CRPS of distributional regression methods. We find the optimal minimax rate of convergence for a given class of distributions and show that the k-nearest neighbor method and the kernel method reach this optimal minimax rate., Comment: Preprint of the article available online in International Journal of Forecasting

Machine learning classification methods usually assume that all possible classes are sufficiently present within the training set. Due to their inherent rarities, extreme events are always under-represented and classifiers tailored for predicting extremes need to be carefully designed to handle this under-representation. In this paper, we address the question of how to assess and compare classifiers with respect to their capacity to capture extreme occurrences. This is also related to the topic of scoring rules used in forecasting literature. In this context, we propose and study a risk function adapted to extremal classifiers. The inferential properties of our empirical risk estimator are derived under the framework of multivariate regular variation and hidden regular variation. A simulation study compares different classifiers and indicates their performance with respect to our risk function. To conclude, we apply our framework to the analysis of extreme river discharges in the Danube river basin. The application compares different predictive algorithms and test their capacity at forecasting river discharges from other river stations.

Heavy rainfall distributional modeling is essential in any impact studies linked to the water cycle, e.g.\ flood risks. Still, statistical analyses that both take into account the temporal and multivariate nature of extreme rainfall are rare, and often, a complex de-clustering step is needed to make extreme rainfall temporally independent. A natural question is how to bypass this de-clustering in a multivariate context. To address this issue, we introduce the stable sums method. Our goal is to incorporate time and space extreme dependencies in the analysis of heavy tails. To reach our goal, we build on large deviations of regularly varying stationary time series. Numerical experiments demonstrate that our novel approach enhances return levels inference in two ways. First, it is robust concerning time dependencies. We implement it alike on independent and dependent observations. In the univariate setting, it improves the accuracy of confidence intervals compared to the main estimators requiring temporal de-clustering. Second, it thoughtfully integrates the spatial dependencies. In simulation, the multivariate stable sums method has a smaller mean squared error than its component-wise implementation. We apply our method to infer high return levels of daily fall precipitation amounts from a national network of weather stations in France.

Accurate estimation of daily rainfall return levels associated with large return periods is needed for a number of hydrological planning purposes, including protective infrastructure, dams, and retention basins. This is especially relevant at small spatial scales. The ERA-5 reanalysis product provides seasonal daily precipitation over Europe on a 0.25 x 0.25 grid (about 27 x 27 km). This translates more than 20,000 land grid points and leads to models with a large number of parameters when estimating return levels. To bypass this abundance of parameters, we build on the regional frequency analysis (RFA), a well-known strategy in statistical hydrology. This approach consists in identifying homogeneous regions, by gathering locations with similar distributions of extremes up to a normalizing factor and developing sparse regional models. In particular, we propose a step-by-step blueprint that leverages a recently developed and fast clustering algorithm to infer return level estimates over large spatial domains. This enables us to produce maps of return level estimates of ERA-5 reanalysis daily precipitation over continental Europe for various return periods and seasons. We discuss limitations and practical challenges and also provide a git hub repository. We show that a relatively parsimonious model with only a spatially varying scale parameter can compete well against statistical models of higher complexity., Comment: 13 pages and 11 figures

A recurrent question in climate risk analysis is determining how climate change will affect heavy precipitation patterns. Dividing the globe into homogeneous sub-regions should improve the modelling of heavy precipitation by inferring common regional distributional parameters. In addition, in the detection and attribution (D&A) field, biases due to model errors in global climate models (GCMs) should be considered to attribute the anthropogenic forcing effect. Within this D&A context, we propose an efficient clustering algorithm that, compared to classical regional frequency analysis (RFA) techniques, is covariate-free and accounts for dependence. It is based on a new non-parametric dissimilarity that combines both the RFA constraint and the pairwise dependence. We derive asymptotic properties of our dissimilarity estimator, and we interpret it for generalised extreme value distributed pairs. As a D&A application, we cluster annual daily precipitation maxima of 16 GCMs from the coupled model intercomparison project. We combine the climatologically consistent subregions identified for all GCMs. This improves the spatial clusters coherence and outperforms methods either based on margins or on dependence. Finally, by comparing the natural forcings partition with the one with all forcings, we assess the impact of anthropogenic forcing on precipitation extreme patterns., Comment: 16 pages, 6 figures, proofs are attached

Some properties of chaotic dynamical systems can be probed through features of recurrences, also called analogs. In practice, analogs are nearest neighbours of the state of a system, taken from a large database called the catalog. Analogs have been used in many atmospheric applications including forecasts, downscaling, predictability estimation, and attribution of extreme events. The distances of the analogs to the target state condition the performances of analog applications. These distances can be viewed as random variables, and their probability distributions can be related to the catalog size and properties of the system at stake. A few studies have focused on the first moments of return time statistics for the best analog, fixing an objective of maximum distance from this analog to the target state. However, for practical use and to reduce estimation variance, applications usually require not just one, but many analogs. In this paper, we evaluate from a theoretical standpoint and with numerical experiments the probability distributions of the $K$-best analog-to-target distances. We show that dimensionality plays a role on the size of the catalog needed to find good analogs, and also on the relative means and variances of the $K$-best analogs. Our results are based on recently developed tools from dynamical systems theory. These findings are illustrated with numerical simulations of a well-known chaotic dynamical system and on 10m-wind reanalysis data in north-west France. A practical application of our derivations for the purpose of objective-based dimension reduction is shown using the same reanalysis data.

Due to complex physical phenomena, the distribution of heavy rainfall events is difficult to model spatially. Physically based numerical models can often provide physically coherent spatial patterns, but may miss some important precipitation features like heavy rainfall intensities. Measurements at ground-based weather stations, however, supply adequate rainfall intensities, but most national weather recording networks are often spatially too sparse to capture rainfall patterns adequately. To bring the best out of these two sources of information, climatologists and hydrologists have been seeking models that can efficiently merge different types of rainfall data. One inherent difficulty is to capture the appropriate multivariate dependence structure among rainfall maxima. For this purpose, multivariate extreme value theory suggests the use of a max-stable process. Such a process can be represented by a max-linear combination of independent copies of a hidden stochastic process weighted by a Poisson point process. In practice, the choice of this hidden process is non-trivial, especially if anisotropy, non-stationarity and nugget effects are present in the spatial data at hand. By coupling forecast ensemble data from the French national weather service (M\'et\'eo-France) with local observations, we construct and compare different types of data driven max-stable processes that are parsimonious in parameters, easy to simulate and capable of reproducing nugget effects and spatial non-stationarities. We also compare our new method with classical approaches from spatial extreme value theory such as Brown-Resnick processes., Comment: 15 pages, 5 figures

Numerical climate models are complex and combine a large number of physical processes. They are key tools in quantifying the relative contribution of potential anthropogenic causes (e.g., the current increase in greenhouse gases) on high impact atmospheric variables like heavy rainfall. These so-called climate extreme event attribution problems are particularly challenging in a multivariate context, that is, when the atmospheric variables are measured on a possibly high-dimensional grid. In this paper, we leverage two statistical theories to assess causality in the context of multivariate extreme event attribution. As we consider an event to be extreme when at least one of the components of the vector of interest is large, extreme-value theory justifies, in an asymptotical sense, a multivariate generalized Pareto distribution to model joint extremes. Under this class of distributions, we derive and study probabilities of necessary and sufficient causation as defined by the counterfactual theory of Pearl. To increase causal evidence, we propose a dimension reduction strategy based on the optimal linear projection that maximizes such causation probabilities. Our approach is tested on simulated examples and applied to weekly winter maxima precipitation outputs of the French CNRM from the recent CMIP6 experiment.

Verifying probabilistic forecasts for extreme events is a highly active research area because popular media and public opinions are naturally focused on extreme events, and biased conclusions are readily made. In this context, classical verification methods tailored for extreme events, such as thresholded and weighted scoring rules, have undesirable properties that cannot be mitigated, and the well-known continuous ranked probability score (CRPS) is no exception. In this paper, we define a formal framework for assessing the behavior of forecast evaluation procedures with respect to extreme events, which we use to demonstrate that assessment based on the expectation of a proper score is not suitable for extremes. Alternatively, we propose studying the properties of the CRPS as a random variable by using extreme value theory to address extreme event verification. An index is introduced to compare calibrated forecasts, which summarizes the ability of probabilistic forecasts for predicting extremes. The strengths and limitations of this method are discussed using both theoretical arguments and simulations.

In recent years several attempts have been made to extend tail modelling towards the modal part of the data. Frigessi et al. (2002) introduced dynamic mixtures of two components with a weight function {\pi} = {\pi}(x) smoothly connecting the bulk and the tail of the distribution. Recently, Naveau et al. (2016) reviewed this topic, and, continuing on the work by Papastathopoulos and Tawn (2013), proposed a statistical model which is in compliance with extreme value theory and allows for a smooth transition between the modal and tail part. Incorporating second order rates of convergence for distributions of peaks over thresholds (POT), Beirlant et al. (2002, 2009) constructed models that can be viewed as special cases from both approaches discussed above. When fitting such second order models it turns out that the bias of the resulting extreme value estimators is significantly reduced compared to the classical tail fits using only the first order tail component based on the Pareto or generalized Pareto fits to peaks over threshold distributions. In this paper we provide novel bias reduced tail fitting techniques, improving upon the classical generalized Pareto (GP) approximation for POTs using the flexible semiparametric GP modelling introduced in Tencaliec et al. (2018). We also revisit and extend the secondorder refined POT approach started in Beirlant et al. (2009) to all max-domains of attraction using flexible semiparametric modelling of the second order component. In this way we relax the classical second order regular variation assumptions.

The classical paradigm of scoring rules is to discriminate between two different forecasts by comparing them with observations. The probability distribution of the observed record is assumed to be perfect as a verification benchmark. In practice, however, observations are almost always tainted by errors and uncertainties. If the yardstick used to compare forecasts is imprecise, one can wonder whether such types of errors may or may not have a strong influence on decisions based on classical scoring rules. We propose a new scoring rule scheme in the context of models that incorporate errors of the verification data. We rely on existing scoring rules and incorporate uncertainty and error of the verification data through a hidden variable and the conditional expectation of scores when they are viewed as a random variable. The proposed scoring framework is compared to scores used in practice, and is expressed in various setups, mainly an additive Gaussian noise model and a multiplicative Gamma noise model. By considering scores as random variables one can access the entire range of their distribution. In particular we illustrate that the commonly used mean score can be a misleading representative of the distribution when this latter is highly skewed or have heavy tails. In a simulation study, through the power of a statistical test and the computation of Wasserstein distances between scores distributions, we demonstrate the ability of the newly proposed score to better discriminate between forecasts when verification data are subject to uncertainty compared with the scores used in practice. Finally, we illustrate the benefit of accounting for the uncertainty of the verification data into the scoring procedure on a dataset of surface wind speed from measurements and numerical model outputs.

Multiple changes in Earth's climate system have been observed over the past decades. Determining how likely each of these changes are to have been caused by human influence, is important for decision making on mitigation and adaptation policy. Here we describe an approach for deriving the probability that anthropogenic forcings have caused a given observed change. The proposed approach is anchored into causal counterfactual theory (Pearl 2009) which has been introduced recently, and was in fact partly used already, in the context of extreme weather event attribution (EA). We argue that these concepts are also relevant, and can be straightforwardly extended to, the context of detection and attribution of long term trends associated to climate change (D&A). For this purpose, and in agreement with the principle of "fingerprinting" applied in the conventional D&A framework, a trajectory of change is converted into an event occurrence defined by maximizing the causal evidence associated to the forcing under scrutiny. Other key assumptions used in the conventional D&A framework, in particular those related to numerical models error, can also be adapted conveniently to this approach. Our proposal thus allows to bridge the conventional framework with the standard causal theory, in an attempt to improve the quantification of causal probabilities. An illustration suggests that our approach is prone to yield a significantly higher estimate of the probability that anthropogenic forcings have caused the observed temperature change, thus supporting more assertive causal claims.

Rainfall ensemble forecasts have to be skillful for both low precipitation and extreme events. We present statistical post-processing methods based on Quantile Regression Forests (QRF) and Gradient Forests (GF) with a parametric extension for heavy-tailed distributions. Our goal is to improve ensemble quality for all types of precipitation events, heavy-tailed included, subject to a good overall performance. Our hybrid proposed methods are applied to daily 51-h forecasts of 6-h accumulated precipitation from 2012 to 2015 over France using the M{\'e}t{\'e}o-France ensemble prediction system called PEARP. They provide calibrated pre-dictive distributions and compete favourably with state-of-the-art methods like Analogs method or Ensemble Model Output Statistics. In particular, hybrid forest-based procedures appear to bring an added value to the forecast of heavy rainfall.

The analysis of seasonal or annual block maxima is of interest in fields such as hydrology, climatology or meteorology. In connection with the celebrated method of block maxima, we study several tests that can be used to assess whether the available series of maxima is identically distributed. It is assumed that block maxima are independent but not necessarily generalized extreme value distributed. The asymptotic null distributions of the test statistics are investigated and the practical computation of approximate p-values is addressed. Extensive Monte-Carlo simulations show the adequate finite-sample behavior of the studied tests for a large number of realistic data generating scenarios. Illustrations on several environmental datasets conclude the work., Comment: 36 pages, 8 tables

We describe a new approach allowing for systematic causal attribution of weather and climate-related events, in near-real time. The method is purposely designed to facilitate its implementation at meteorological centers by relying on data treatments that are routinely performed when numerically forecasting the weather. Namely, we show that causal attribution can be obtained as a by-product of so-called data assimilation procedures that are run on a daily basis to update the meteorological model with new atmospheric observations; hence, the proposed methodology can take advantage of the powerful computational and observational capacity of weather forecasting centers. We explain the theoretical rationale of this approach and sketch the most prominent features of a "data assimilation-based detection and attribution" (DADA) procedure. The proposal is illustrated in the context of the classical three-variable Lorenz model with additional forcing. Several theoretical and practical research questions that need to be addressed to make the proposal readily operational within weather forecasting centers are finally laid out.