Machine learning techniques to derive bioclimatic classifications for Colombia

Abstract

Bioclimatic classifications seek to divide a study region into geographic areas with similar bioclimatic characteristics. In this study we proposed two bioclimatic classifications for Colombia using machine learning techniques. We firstly characterized the precipitation space of Colombia using principal component analysis. Based on Lang classification, we then projected all background sites in the precipitation space with their corresponding categories. We sequentially fit logistic regression models to reclassify all background sites in the precipitation space with six redefined Lang categories. New categories were the used to define a new modified Lang and Caldas-Lang classifications.