Clustering for hydroclimatic extremes: a retrospective synthesis of the literature

Earth system processes have complex physics and are dynamically interlinked, making modelling and predictions difficult. In particular, current challenges for hydroclimatic systems are in understanding nonstationarity and heterogeneity driven by climatic and human influences. Hence, studying the spatial and temporal occurrences and dependencies of hydroclimatic extremes is becoming increasingly important for water resources management and hydrological services. 

Research efforts that address these issues for extreme hydrological events are large and diverse in their approaches and methodologies. In this respect, multivariate statistical methods such as clustering are approaches commonly used to reveal mechanisms affecting floods and droughts in relation to their trends and magnitude as well as their variability in time and space. Clustering is a convenient tool to analyze large hydrometeorological datasets because of its unsupervised nature. However, there are no structured insights for hydrologists to reflect on the principles and findings of data clustering for hydroclimatic extremes. 

This contribution sheds light on the why’s and how’s of clustering methods for floods and droughts based on a systematic review of the literature. Our aim is to provide a synthesis of hydrological themes and methodological concepts found in papers that investigate floods and droughts through data clustering. These insights are valuable for guiding future applications of clustering methods while enabling wider discussions on the knowledge gaps for modelling extremes in hydroclimatic systems.