Cai Yi Ru
Research Mentor(s): Dani Jones
Mentor Department: Cooperative Institute for Great Lakes Research (CIGLR)
Authors: Linda Ru, Dani Jones
Session: Session 5 (2:00pm – 2:50pm)
Presentation Type: Poster 3
Abstract
Understanding the impacts of Extratropical Cyclones (ETCs) on a region’s weather patterns, affecting winds, heat transport, precipitation, and evaporation is important for infrastructure management, planning, and improving predictive hydrologic models. Many residents in the US and Canada rely on natural resources for drinking water, commerce, and recreation near the Great Lakes Basin. Our study of ETCs’ impact on the region’s weather patterns can further quantify the differences in impact which also can lead to better predictions of the impacts of the ETCs on the Great Lakes Region. We explored unsupervised classification methods for the impacts of extratropical cyclones on the Great Lakes Basin using 2 different datasets, ERA5 and Climate Forecast System Reanalysis (CFSR). A few different clustering methods – hierarchical clustering, Gaussian Mixture Models clustering (GMM), and K-Means clustering were implemented on the ERA5 dataset to classify and find different storm types, to compare how they impact the weather patterns in different ways. We discovered that the optimal number of clusters is approximately 4-6. While there is some flexibility in choosing the exact number of clusters, it’s important to consider certain constraints: selecting too many clusters may lead to overfitting; conversely, selecting too few clusters might result in a model that does not accurately represent the covariance between features. By using a fixed number of clusters (4) we found that there were differences in impact between the clusters. Specifically, there was one class that seemed to have low/weak impacts on precipitation. Further visualizations suggest that the class with low impacts were also storms that were more mature. This shows that maturity of storms is an important factor in determining the clusters.
