The Solar System Notification Alert Processing System (SNAPS): Asteroid Population Outlier Detection

TL;DR

SNAPS is a system that detects outliers among Solar System objects, particularly asteroids, using unsupervised methods and large-scale data, enabling new scientific investigations.

Contribution

The paper introduces four novel outlier metrics and a scalable approach for population outlier detection in Solar System objects using multiple feature spaces.

Findings

Outlier scores can be computed efficiently at Rubin Observatory scale.

Multiple feature space permutations enhance outlier detection robustness.

SNAPS enables new scientific investigations through population outlier analysis.

Abstract

The Solar System Notification Alert Processing System (SNAPS) is a ZTF and Rubin Observatory alert broker that will send alerts to the community regarding interesting events in the Solar System. SNAPS is actively monitoring Solar System objects and one of its functions is to compare objects (primarily main belt asteroids) to one another to find those that are outliers relative to the population. In this paper, we use the SNAPShot1 dataset which contains 31,693 objects from ZTF and derive outlier scores for each of these objects. SNAPS employs an unsupervised approach; consequently, to derive outlier rankings for each object, we propose four different outlier metrics such that we can explore variants of outlier scores and add confidence to outlier rankings. We also provide outlier scores for each object in each permutation of 15 feature spaces, between 2 and 15 features, which yields 32,752 total feature spaces. We show that we can derive population outlier rankings each month at Rubin Observatory scale using four Nvidia A100 GPUs, and present several avenues of scientific investigation that can be explored using population outlier detection.