Automatic detection of the electron density from de WHISPER instrument onboard CLUSTER II

The Waves of HIgh frequency and Sounder for Probing Electron density by Relaxation
(WHISPER) instrument, is part of the Wave Experiment Consortium (WEC) of the ESA
CLUSTER II mission. WHISPER is designed to measure the electric field fluctuation and derive the electron density, i.e. the plasma density, a key parameter of scientific interest for
magnetospheric and near-Earth solar wind studies. The electron density is the WHISPER highest level product and is provided, among other products, to the scientific community through the CLUSTER Science Archive (CSA).
The instrument consists of a receiver, a transmitter, and a wave spectrum analyzer. It delivers both ambient (in natural mode) and active (in sounding mode) electric field spectra. The characteristic signatures of ambient plasma waves or active plasma resonances, combined with the spacecraft position, reveal the different magnetosphere regions. These spectral signatures are used to derive the electron density. Until recently, ad-hoc algorithms have been used to derive the electron density from WHISPER measurements, but at the cost of time-consuming manual steps. These algorithms are dependent on measurements provided by other instruments onboard CLUSTER, thus introducing dependencies and potential delays in the data production.

In this context, the goal of this work is to significantly reduce human intervention by fully
automating the WHISPER electron density derivation, exclusively using WHISPER data.
For this purpose, we develop a two-step derivation process, based on neural networks: first, the plasma region is identified with a Multi-Layer Perceptron classification algorithm; second, the electron density is derived using a Recurrent Neural Network, adapted to each plasma region. These networks have been trained with WHISPER spectra and electron density previously derived from ad-hoc algorithms. The resulting accuracy is up to 98% in some plasma regions. This derivation process has been implemented in a production pipeline, now routinely used to deliver WHISPER electron density to the CSA and dividing by 10 the human intervention. The pipeline has already delivered 3+ years of data and will be used to reprocess some of the archive focusing on the most complex plasma regions with recent improvements. This work will present the implemented methods and models for each region focusing on results and performance.