Articles | Volume 8, issue 3
Earth Syst. Dynam., 8, 677–696, 2017
Earth Syst. Dynam., 8, 677–696, 2017

Research article 08 Aug 2017

Research article | 08 Aug 2017

Multivariate anomaly detection for Earth observations: a comparison of algorithms and feature extraction techniques

Milan Flach1, Fabian Gans1, Alexander Brenning2,4, Joachim Denzler3,4,5, Markus Reichstein1,4,5, Erik Rodner3,4, Sebastian Bathiany6, Paul Bodesheim1, Yanira Guanche3,4, Sebastian Sippel1, and Miguel D. Mahecha1,4,5 Milan Flach et al.
  • 1Max Planck Institute for Biogeochemistry, Department Biogeochemical Integration, P.O. Box 10 01 64, 07701 Jena, Germany
  • 2Friedrich Schiller University Jena, Department of Geography, Jena, Germany
  • 3Friedrich Schiller University of Jena, Department of Mathematics and Computer Sciences, Computer Vision Group, Jena, Germany
  • 4Michael Stifel Center Jena for Data-driven and Simulation Science, Jena, Germany
  • 5German Centre for Integrative Biodiversity Research (iDiv), Leipzig, Germany
  • 6Wageningen University, Department of Environmental Sciences, Wageningen, the Netherlands

Abstract. Today, many processes at the Earth's surface are constantly monitored by multiple data streams. These observations have become central to advancing our understanding of vegetation dynamics in response to climate or land use change. Another set of important applications is monitoring effects of extreme climatic events, other disturbances such as fires, or abrupt land transitions. One important methodological question is how to reliably detect anomalies in an automated and generic way within multivariate data streams, which typically vary seasonally and are interconnected across variables. Although many algorithms have been proposed for detecting anomalies in multivariate data, only a few have been investigated in the context of Earth system science applications. In this study, we systematically combine and compare feature extraction and anomaly detection algorithms for detecting anomalous events. Our aim is to identify suitable workflows for automatically detecting anomalous patterns in multivariate Earth system data streams. We rely on artificial data that mimic typical properties and anomalies in multivariate spatiotemporal Earth observations like sudden changes in basic characteristics of time series such as the sample mean, the variance, changes in the cycle amplitude, and trends. This artificial experiment is needed as there is no gold standard for the identification of anomalies in real Earth observations. Our results show that a well-chosen feature extraction step (e.g., subtracting seasonal cycles, or dimensionality reduction) is more important than the choice of a particular anomaly detection algorithm. Nevertheless, we identify three detection algorithms (k-nearest neighbors mean distance, kernel density estimation, a recurrence approach) and their combinations (ensembles) that outperform other multivariate approaches as well as univariate extreme-event detection methods. Our results therefore provide an effective workflow to automatically detect anomalies in Earth system science data.

Short summary
Anomalies and extremes are often detected using univariate peak-over-threshold approaches in the geoscience community. The Earth system is highly multivariate. We compare eight multivariate anomaly detection algorithms and combinations of data preprocessing. We identify three anomaly detection algorithms that outperform univariate extreme event detection approaches. The workflows have the potential to reveal novelties in data. Remarks on their application to real Earth observations are provided.
Final-revised paper