Zoomed-in burst-envelope forecasts from January 2024 to April 2026. Top: The Northern Hemisphere SSN with inferred RF (Random Forest) bursts properties. Middle: The Southern hemisphere counterpart. Bottom: Total (Northern Hemisphere + Southern Hemisphere) SSN, highlighting the same burst timings and amplitudes.
The Astrophysical Journal: A novel machine-learning-based hybrid forecasting strategy for predicting next enhanced burst of solar activity is presented. This hybrid forecast-system combines numerical, statistical, and machine-learning techniques to detect the occurrence of the next bursts of solar activity. These enhanced bursts are called “space weather seasons” that occur on intermediate timescales (6–18 months). Monthly smoothed sunspot number (SSN) data from 1878 to 2025 are analyzed using Gaussian fitting techniques to identify burst events and their properties such as amplitude and duration. The SSN data are divided into training, test, and forecast, which shows hindcast and forecast. Each hemisphere is modeled via a Seasonal Autoregressive Integrated Moving Average (SARIMA) approach, refined with an asymmetric Gaussian override to capture rapid burst rise and gradual decay, and burst amplitudes and duration are predicted using a Random Forest (RF) regression model. This hybrid approach successfully hindcasts burst timing in between November 2024 and May 2025, with a peak SSN of ∼70 around March 2025 for the Northern Hemisphere. The next burst in the Northern Hemisphere is forecast to be in December 2025 with a slightly lower SSN of 60. By contrast, the Southern Hemisphere shows relatively complicated behavior, where the bursts show multiple amplitudes starting approximately in October 2024 and ending in October 2025. The main burst shows an amplitude of 130 SSN. The next burst in the Southern Hemisphere is forecast to occur approximately in December 2025. Combining SSN properties in both hemispheres, we find that the total SSN is mainly influenced by a stronger cycle in the Southern Hemisphere.