Climate regionalization is an inseparable part of many climate change and environmental studies. Delineating climatologically homogeneous regions enhances the utility of such studies and reduces the biases due to the uncertainties associated with climate model outputs at individual grid points which both lead to better understanding of the atmospheric mechanisms affecting a region's climate. Throughout time, researchers and statisticians have developed different methods to perform regionalization in which the techniques are highly dependent on the nature and accessibility of the data. This research aims to divide Bolivia into smaller, coherent climate subdivisions. To achieve this goal, we first apply the non-hierarchical k-means clustering method to climatologies of monthly accumulated precipitation and monthly average temperature separately using a gridded observation dataset for Bolivia spanning from 1979 to 2010. The clustering is performed on the two variables separately to avoid arbitrary attribute scaling and information redundancy as well as to gain a better understanding of these individual variables across Bolivia. Consensus clustering then finds the categorical intersection of the two independent clusters to create homogeneous climate regions. Results from this study show that Bolivia can be divided into 10 climatically distinguishable subdivisions largely explicable by topography and latitude, which are the key climate control factors in the region.
Bibliographical noteFunding Information:
We gratefully acknowledge support from the Interamerican Development Bank for the development of tools and techniques used in this research and the UNL Holland Computing Center for computing services and support.
© 2019 Royal Meteorological Society
- climate regionalization
- consensus clustering
- k-means clustering