Automated outlier detection with machine learning in GRACE and GRACE-FO post-fit residuals

Zbinden, Jonas; Lasser, Martin; Meyer, Ulrich; Panos, Brandon; Arnold, Daniel; Jäggi, Adrian (July 2023). Automated outlier detection with machine learning in GRACE and GRACE-FO post-fit residuals (Unpublished). In: 28th IUGG General Assembly 2023. Berlin, Germany. 11--20 July 2023.

Preview

Text poster-iugg2023_Zbinden.pdf - Other Available under License BORIS Standard License. Download (11MB) | Preview

GRACE/GRACE-FO inter-satellite range-rates are the main observable for the determination of the monthly solutions of the Earthâ?Ts gravity field. The range-rates are sensitive not only to the mass distribution on the Earth, and as a consequence, the relative motion of both GRACE and GRACE-FO satellites, respectively, but also to the relative orientation of the satellites and consequently on the attitude handling, which relies on the star camera observations, object to blinding by sun and moon. In consequence the range-rates observations exhibit a number of difficult to identify error sources and efficient screening is not trivial. We therefore apply novel outlier detection methods such as Isolation Forest and Local Outlier Factor, (collectively known under the term machine learning), to flag outliers in an unsupervised fully automated way. We apply both techniques to the post-fit residuals of monthly, joined orbit and gravity field determination processes, combined with the geographical position of each observation. The flagged outliers are investigated for local geographical correlations to distinguish between unfitted signal from graivitational sources and artefacts caused by the satellite's instrumentation. For that we train a Mutual Information Neural Network, learning the mutual information between the post-fit residuals and the geographical location. The outliers flagged as artefacts are removed from the original inter-satellite range-rate data and orbit and gravity field determination process is repeated to investigate if the gravitational model fits improves.

Interest & Impact

Downloads

21 since deposited on 21 Aug 2023
11 in the past 12 months

Citations

5 Citations in Web of Science ®
6 Citations in Scopus

Search

in Google Scholar™

Services

Actions (login required)

Edit item

Actions (login required)

Edit item