Aurora Hunter: A Two-Stage Framework for Probabilistic Visibility Forecasting

2026-05-21

AI Summary

This study addresses the challenge of aurora visibility prediction, which requires simultaneous modeling of both auroral physical occurrence and observational conditions such as cloud cover and moonlight—factors that traditional methods struggle to integrate effectively. To this end, the authors propose Aurora Hunter, a two-stage cascaded framework: the first stage employs XGBoost to predict the probability of auroral occurrence, while the second stage uses logistic regression to estimate the conditional probability of clear observation given that an aurora occurs; the final visibility score is the product of these two probabilities. This approach uniquely decouples visibility into independent physical and observational components, leveraging 51 physics-driven features and 21 meteorological/astronomical features, with SHAP analysis enhancing interpretability. Evaluated on test sets from Tromsø and Kiruna, the model achieves ROC-AUC scores of 0.937 and 0.905, respectively—outperforming single-stage baselines by 0.087—and demonstrates strong cross-site generalization on Skibotn data.

Abstract

Forecasting aurora borealis visibility matters for space weather research and aurora tourism. Visibility at a site and night depends on two distinct factors: (1) whether aurora is physically occurring, driven by solar wind-magnetosphere coupling, and (2) whether observing conditions allow naked-eye detection, mainly cloud cover and lunar illumination. We present Aurora Hunter, a two-stage cascade that decouples these factors. Stage 1 predicts P(occurring) with XGBoost using 51 physics-driven features trained on joint Tromso+Kiruna data (about 16,600 hourly samples, 2015-2023) with labels from the Tromso AI all-sky image classifier. Stage 2 predicts P(clear observation given occurring) with logistic regression using 21 cloud-cover and lunar-illumination features trained only on aurora-occurring hours. The cascade P(visible)=P(occurring)*P(clear|occurring) reaches ROC-AUC 0.937 (Tromso test, 2019-2020) and 0.905 (independent Kiruna, 2024), improving a single-stage baseline by +0.087. Held-out Skibotn data (2022-2025) confirm cross-site generalization. SHAP identifies the Kp x nightside interaction, MLT position, and auroral oval distance as dominant predictors (39% combined). Prototype: https://aurora-hunter.onrender.com.

Problem

• aurora visibility
• space weather
• cloud cover
• lunar illumination
• solar wind-magnetosphere coupling

Innovation

• two-stage cascade
• probabilistic visibility forecasting
• aurora borealis prediction
• physics-driven features
• cross-site generalization