A solar spectral irradiance prediction workflow using a recurrent neural network in a reproducible and replicable approach

The solar irradiance measured at the top of the Earth’s atmosphere is considered an influential parameter in studies of the atmosphere layers’ properties and the consequences of the disturbances they suffer from the influence of solar activity. Studies about weather and climate on Earth recognized the great influence of solar irradiation for the creation of climate models, so the prediction of solar irradiance can be considered a service of great importance in this context. An irradiance prediction system would also be useful for the reconstruction of the measurement history (time series) of different instruments that do not cover the intended period data, due to reasons like mismatches in calibration or failures on the instruments. This work uses recurrent neural network to predict solar spectral irradiance in different wavelengths. In the experiments, one and a half year’ daily records collected by the SDO (Solar Dynamics Observatory) mission are used. Two types of images of the solar photosphere, collected by the HMI (Helioseismic and Magnetic Imager) equipment, which highlight active regions and sunspots, are processed and used as the network input. For the neural network output, spectral irradiance data, collected by the EVE equipment, at two different lines of helium, at 30.5nm and 48.5 nm, and a line of hydrogen, the lyman alpha at 121.6 nm, are used. In addition to the forecasting task, this work is also an early step in the adoption of a modular workflow proposal that allows its fully validation, reuse and replication.


Theme – Machine Learning, Statistics, and Algorithms