Trajectory Clustering Based Oceanic Anomaly Detection Using Argo Profile Floats

Abstract

The observation data of Argo profile floats are very crucial for long-term climate change and natural variability, which reflect three-dimensional distribution of temperature and salinity in the sea. In order to solve the anomalies in the profile caused by uncertainties factors, this paper proposes a novel anomaly detection method for Argo profile floats using an improved trajectory clustering method to discriminate normal and abnormal. The proposed algorithm partitions Argo data into a set of line segments, and then clusters line segments to get rid of noisy data, finally recovers the line segments to the raw data accordingly. As a result, the proposed oceanic anomaly detection method subtly converts the sequence data into line segments for anomaly detection, which considers both positional relationship and trend of data source. Extensive experiments on real dataset from Argo floats verify that our method has better results under different conditions compared to existing methods such as LOF and DBSCAN.