Transformed-stationary EVA 2.0: a generalized framework for non-stationary multivariate extremes analysis

The increasing availability of extensive time series on natural hazards underscores the need for robust non-stationary methods to analyze evolving extremes. Moreover, growing evidence suggests that jointly analyzing phenomena traditionally treated as independent, such as storm surge and river discharge, is crucial for accurate hazard assessment. While univariate non-stationary extreme value analysis (EVA) has seen substantial development in recent decades, a comprehensive methodology for addressing non-stationarity in joint extremes – compound events involving simultaneous extremes in multiple variables – is still lacking. To fill this gap, here we propose a general framework for the non-stationary analysis of joint extremes that combines the Transformed-Stationary Extreme Value Analysis (tsEVA) approach with Copula theory. This methodology implements sampling techniques to extract joint extremes, applies tsEVA to estimate non-stationary marginal distributions using GEV or GPD distributions, and utilizes time-dependent copulas to model evolving inter-variable dependencies. The approach's versatility is demonstrated through case studies analyzing historical time series of significant wave height, river discharge, temperature, and drought, uncovering dynamic dependency patterns over time. To support broader adoption, we provide an open-source MATLAB toolbox that implements the methodology, complete with examples, available on GitHub.