Abstract

Satellite imagery, such as from Sentinel 1 and Sentinel 2 satellites, has demonstrated reliable accuracy in identifying agricultural phenomena like harvest times and crop rotation types. These patterns are discerned through sequential analysis of satellite image time series across specific periods. However, the sequential patterns in the time series data are often coarse, and anomalies can disrupt further analysis. Detecting anomalies in time series data is particularly challenging, as they can manifest as anomalies at single time points or anomalies across entire sequences.

To address these challenges, recent studies have explored advanced algorithms, particularly clustering algorithms applied to time series data. Dimensional Time Warping (DTW) is a commonly used distance measurement method for sequential data due to its effectiveness in various applications. By utilizing DTW to calculate dissimilarity matrices from the satellite image time series, researchers can conduct anomaly detection using clustering approaches.

This paper focuses on analyzing anomalies derived from satellite image time series using a clustering approach with DTW as the dissimilarity measurement. Specifically, it employs DBscan clustering after computing dissimilarity matrices. To enhance the analysis, the paper compares the results obtained directly from the dissimilarity matrices with those derived from low-dimensional representations using Self Organizing Map (SOM) and Multidimensional Scaling (MDS) approaches.

The study applies these anomaly detection techniques to monitor crop anomalies in West Java, Indonesia, integrating data from both Sentinel 1 and Sentinel 2 satellites. This research aims to improve the accuracy and reliability of anomaly detection in agricultural monitoring using advanced satellite image analysis techniques.