Speaker
Description
One common feature of many fields is the necessity to forecast the temporal evolution of signals and the question is always more challenging when the signal is irregular and multiscale. For instance, we can find forecasting the temperature variation or the localisation of rainy clouds in meteorology, the daily or monthly stock exchange for finance applications, tiny variations in electrocardiograms to diagnose cardiac ailments, or forecasting the solar, solar wind, and geomagnetic activity to prevent threats to our society from space weather.
The Analogue Ensemble technique originates from Lorentz’s idea, to forecast the weather, that we can forecast the future evolution by finding similar occurrences of the present behaviour in the past and by considering the progression of these occurrences. This idea leads to an ensemble of forecasts. We use this technique to forecast the magnetic and velocity field of the solar wind for Space Weather applications. By comparing the obtained forecasts to climatology (long-term average), persistence (future constant progression of the present state), and recurrence (cyclic occurrence), we have demonstrated that the fluctuations of these quantities can be well estimated with the AnEn method. We can describe, scale by scale, the fluctuations of a signal with a spectrum in the Fourier frequency space. As a multiscale system, the spectra of solar wind quantities span a broad range of frequencies and should be reflected in the forecast. We propose novel ensemble-reduction and diagnostics techniques to preserve and measure the spectral performance of the ensemble forecast.
