Solar Cycle Key Node Prediction Using NodeFusion

The solar cycle (SC) is an important timescale for describing the long-term evolution of solar activity, and understanding its evolutionary patterns is of great significance for space weather research and related applications. However, due to the complex nonlinear nature of solar activity evolution, accurately predicting SC evolution remains challenging. Existing research has largely focused on predicting sunspot number (SSN) time series and cycle peaks, while systematic studies of multiple key nodes of a cycle remain relatively limited. This paper proposes NodeFusion, a multi-expert fusion method for predicting four key nodes of a cycle: start time, peak time, end time, and peak amplitude. The method integrates multiple heterogeneous forecasting experts to generate candidate predictions and derives the final predictions through node-level fusion and temporal calibration based on available historical forecast errors. The study uses monthly SSN observations provided by the World Data Center for Sunspot Index and Long-term Solar Observations (SILSO), Version 2.0. NodeFusion is evaluated under a strict issue-time rolling-origin protocol on SC23–SC25, with the end-time evaluation restricted to SC23–SC24 because SC25 has not yet ended. The results show that, compared with the best-performing individual expert at each node, NodeFusion achieves mean absolute error (MAE) reductions of 6.7 per cent, 21.4 per cent, 14.1 per cent, and 50.4 per cent for the start time, peak time, end time, and peak amplitude, respectively. Using observations available through June 2026, NodeFusion predicts that SC26 will begin in October 2030, reach its peak in November 2035 with a peak amplitude of 116.9, and end in March 2042; the associated forecast uncertainty is quantified using residual-based 95% uncertainty ranges.