Dresden 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 6: Machine Learning in Dynamics and Statistical Physics I
DY 6.7: Vortrag
Montag, 9. März 2026, 11:15–11:30, HÜL/S186
Checking the superiority of multi-model mean forecasts by reservoir computing — Daniel Estevez Moya1,3, Erick A. Madrigal Solis1,2, Ernesto Estevez Rams3, and •Holger Kantz1 — 1Max Planck Institute for the Physics of Complex Systems, Dresden, Germany — 2University of Techology, Dresden, Germany — 3Faculdad de Física, Universidad de La Habana, Cuba
In weather prediction and climate forecasts it has been observed that taking the arithmetic mean forecast of an ensemble of different models is often superior to most of the individual models. We use Reservoir Computing to generate easily a large ensemble of models and study their performance on deterministic toy models. While each individually trained reservoir comes with its own model error which is a systematic error, we verify that the arithmetic mean of these forecasts is closer to the truth than most of the individual forecasts. We present a detailed dynamical explanation for this observation.
Keywords: ensemble forecasts; reservoir computing