FLARE: A Framework for Stellar Flare Forecasting using Stellar Physical Properties and Historical Records

TL;DR

Stellar flare forecasting is challenged by scarce flare samples. The authors propose FLARE, a large multi-modal framework that predicts whether a flare occurs within the next $H=24$ hours by integrating light-curve data with stellar physical properties and historical flare records. FLARE combines a Light Curve Embedding with a Soft Prompt Module for properties and a Residual Record Fusion Module for historical flares, and uses LoRA fine-tuning of a pre-trained large model to produce a probability $\hat{y}_{(t,t')}^i$ under a cross-entropy loss with label smoothing. On the Kepler dataset, FLARE achieves superior performance across metrics and ablation studies validate the contributions of both SPPs and HFRs, as well as the effectiveness of the prompt and fusion components. This work demonstrates the practical value of cross-modal temporal forecasting in astrophysics and provides a blueprint for expanding stellar activity datasets.

Abstract

Stellar flare events are critical observational samples for astronomical research; however, recorded flare events remain limited. Stellar flare forecasting can provide additional flare event samples to support research efforts. Despite this potential, no specialized models for stellar flare forecasting have been proposed to date. In this paper, we present extensive experimental evidence demonstrating that both stellar physical properties and historical flare records are valuable inputs for flare forecasting tasks. We then introduce FLARE (Forecasting Light-curve-based Astronomical Records via features Ensemble), the first-of-its-kind large model specifically designed for stellar flare forecasting. FLARE integrates stellar physical properties and historical flare records through a novel Soft Prompt Module and Residual Record Fusion Module. Our experiments on the publicly available Kepler light curve dataset demonstrate that FLARE achieves superior performance compared to other methods across all evaluation metrics. Finally, we validate the forecast capability of our model through a comprehensive case study.

Figures (7)

• Figure 1: Star observations in multiple Extreme Ultraviolet (EUV) wavelengths before, during, and after a stellar flare.
• Figure 2: Light curves of several stars: (a) The flare region exhibits greater intensity in flux variations compared to non-flare regions. (b) The same star at different times displays distinct fluctuation patterns in its light curve. (c) Different stars during the same observation period display notable variations in their light curves, highlighting diversity across stellar systems.
• Figure 3: The overall framework of FLARE. First, the light curve is decomposed into trend and residual components, which are processed separately through patching and the Residual Record Fusion Module integrated with flare records. Timestamp embeddings are then appended to these processed components. Simultaneously, stellar physical properties are embedded using the Soft Prompt Module, generating a corresponding vector representation. The resulting vectors from both pathways are concatenated and passed through a large model. Finally, an MLP head processes the output to predict the probability of flare occurrence or non-flare conditions within the next 24 hours.
• Figure 4: (a) A textual description of the star KIC 011924842's physical properties. (b) Two replacement pattens and examples.
• Figure 5: FLARE forecasts whether flares will occur in multiple samples. The purple and green region represent the observation and the forecast area, and red dots mark the time steps that belong to the flares. "forecast flare occurrence" and "forecast non-flare period" are used to represent the forecasting results of FLARE.