Abstract
Nowadays, massive climate simulation data sets are produced due to the unprecedented increase in computing power, and there is a need to provide automated methods for analyzing these data. Here we focus on one particular class of methods, that is, methods for local detection of extreme weather phenomena. We describe an automated method for the identification of the extreme events in large sets of climate simulation data. This method adapts an algorithm for topological data analysis to extract numerical features of topological descriptors called connected components. The features are then fed to a supervised machine learning classifier. The classifier performs a binary classification task to identify the extreme weather patterns we are interested in. We illustrate capabilities of this method by presenting a case study of atmospheric river patterns that are often associated with severe precipitation in the mid-latitudes. We also show that the method can be suitable for analyzing large amounts of climate simulation products. Hence, we think that the climate community will find this example instructive and inspiring. We also indicate other future climate science problems in which applied topology coupled with machine learning can be found useful.