📝 SummarXiv

Abstract

Accurate prediction of typhoon trajectories is essential for mitigating the impact of these extreme weather events. This study proposes a functional data analysis (FDA) framework for modeling and forecasting typhoon paths using historical trajectory data. Latitude and longitude sequences are represented as smooth functional observations, and a function-on-function regression model is employed to capture the temporal dynamics of typhoon movement. While the baseline model demonstrates reasonable performance for typical bow-shaped trajectories, it exhibits limited accuracy for non-standard paths. To address this limitation, a clustering-based extension is introduced, wherein typhoons are grouped by trajectory shape prior to regression. This two-stage approach improves predictive accuracy by enabling localized modeling adapted to structural variations in the data. The results demonstrate the practical utility of combining FDA with clustering for robust and flexible typhoon trajectory forecasting.