A regionalisation approach for rainfall based on extremal dependence (Q2028583)

In modelling extreme rainfall events, the article presents a regionalization method that partitions stations into regions of similar extremal dependence using hierarchical clustering with the F-madogram distance and a classification step using a weighted \(k\)-nearest neighbour classifier. To demonstrate the approach, the article considers a study region of a network of daily rainfall stations in Australia and discusses, in detail, the results with respect to known climate and topographic features. Max-stable models are fitted to the observed annual maximal rainfall in each of the regions to visualize and evaluate the effectiveness of the partitioning relative to the full extremal dependence.