Forecast of solar activity based on mean-field dynamo model and neural network

We discuss a prediction of the solar activity on a short time-scale applying the method based on a combination of a nonlinear mean-field dynamo model and the artificial neural network. The artificial neural network which serves as a correction scheme for the forecast, uses the currently available observational data (e.g., the 13 month running average of the observed solar sunspot numbers) and the dynamo model output. The nonlinear mean-field \(\alpha\,\Omega\) dynamo produces the large-scale magnetic flux which is redistributed by negative effective magnetic pressure instability (NEMPI) producing sunspots and active regions. The nonlinear mean-field dynamo model includes algebraic nonlinearity (caused by the feedback of the growing magnetic field on the plasma motion) and dynamic nonlinearities (related to the dynamics of the magnetic helicity of small-scale magnetic field). We compare the forecast errors with a horizon of 1, 6, 12 and 18 months, for different forecast methods, with the same corrections on the current monthly observations. Our forecast is in good agreement with the observed solar activity, the forecast error is almost stably small over short-medium ranges of forecasting windows. Despite a strong level of chaotic component in the solar magnetic activity we present quantitative evidence that the solar activity on a short range can be stably well predicted, by the joint use of the physically based model with the neural network. This result may have an immediate practical implementation for predictions of various phenomena of solar activity and other astrophysical processes, so may be of interest to a broad community.