World climate search and classification using a dynamic time warping similarity function

Abstract

We present a data-mining approach to climate classification and analysis. Local climates are represented as time series of climatic variables. A similarity between two local climates is calculated using the dynamic time warping (DTW) function that allows for scaling and shifting of the time axis to model the similarity more appropriately than a Euclidean function. A global grid of climatic data is clustered into 5 and 13 climatic classes, and the resultant world-wide map of climate types is compared to the empirical Köppen–Geiger classification. We also present a concept of climate search—an interactive, Internet-based application that allows retrieval and mapping of world-wide locations having climates similar to a user selected location query.