Extracting patterns from multivariate weather data

Summary

Weather data include measurements featuring complex spatiotemporal relations and patterns that cannot be described adequately by linear solutions. Working on such data is even more challenging, as no traditional supervised techniques or extrinsic evaluation of a clustering outcome can be applied due to the absence of labels or annotations. Initially, we will explore and evaluate segmentation algorithms for multivariate weather time series over a target area, setting a foundation for further processing. We will then apply machine learning and deep learning on the resulting segments, for extracting weather patterns. We will seek to evaluate our findings using qualitative and/ or quantitative techniques, at the same time emphasizing the reusability of our methods and data products.