Climate Regionalization through Hierarchical Clustering: Options and Recommendations for Africa
Friday, 19 December 2014
Climate regionalization is an important but often under-emphasized step in studies of climate variability and predictions. While most investigations of regional climate or statistical/dynamical predictions do make at least an implicit attempt to focus on a study region or sub-regions that are climatically coherent in some respect, rigorous climate regionalization––in which the study area is divided on the basis of the most relevant climate metrics and at a resolution most appropriate to the data and the scientific question––has the potential to enhance the precision and explanatory power of climate studies in many cases. This is particularly true for climatically complex regions such as the Greater Horn of Africa (GHA) and Equatorial West Africa. Here we present an improved clustering method and a flexible, open-source software tool (R package “HiClimR”) designed specifically for climate regionalization. As a demonstration, we apply HiClimR to regionalize the GHA on the basis of interannual precipitation variability in each calendar month and for three-month running seasons. Different clustering methods are tested to show the behavior of each method and provide recommendations for specific problems. This would underscore the applicability of our work to a wide range of climate issues, and enable researchers to easily and quickly learn how to apply our tools to their own problems. Both the proposed methodology and the R package can be easily used for a broad range of climate applications.