LAUSR

Abstract. Now that the Earth has been monitored by satellites for more than 40 years, Earth observation images can be used to study how the Earth system behaves over extended periods. Such long-term studies require the combination of data from multiple instruments, with the earliest datasets being of particular importance in establishing a baseline for trend analysis. As the quality of these earlier datasets is often lower, careful quality control is essential, but the sheer size of these image sets makes an inspection by hand impracticable. Therefore, one needs to resort to automatic methods to inspect these Earth observation images for anomalies. In this paper, we describe the design of a system that performs an automatic anomaly analysis on Earth observation images, in particular the Meteosat First Generation measurements. The design of this system is based on a preliminary analysis of the typical anomalies that can be found in the dataset. This preliminary analysis was conducted by hand on a representative subset and resulted in a finite list of anomalies that needed to be detected in the whole dataset. The automated anomaly detection system employs a dedicated detection algorithm for each of these anomalies. The result is a system with a high probability of detection and low false alarm rate. Furthermore, most of these algorithms are able to pinpoint the anomalies to the specific pixels affected in the image, allowing the maximum use of the data available.