A central activity in T-FORS project is the development and validation of new models capable of providing TID forecasts.
Two comolementary models provide Large Scale Travelling Ionospheric Disturbances (LSTID) forecasts with a forecasting horizon up to three hours. The models tackle the complexity of physical processes in various regions of the atmosphere, ionosphere, thermosphere, magnetosphere and heliosphere, and provide forecasting with different time horizons. Figure 1 reports the conceptual flow and datasets describing the chain of the physical processes involved in the occurrence of LSTIDs. The two different models ares:
a. Catalogue-based forecasting (3-hrs forecasting horizon), based on the catalogue of LSTID events provided by the HF Interferometry Method and on the exploitation of CatBoost classifiers;
b. LSTID forecasting (2-h forecasting horizon) over Digisonde locations, based on the Spectral Energy Contribution (SEC) index that it is provided by the HF Interferometry Method and on the exploitation of Temporal Fusion Transformers (TFT) classifiers.
The validation includes statistical evaluation, explainable artificial intelligence approaches, and scientific analysis of case events. The validation results provided an improved scientific understanding on the triggering and propagation of LSTIDs, resulting also in an inventory of early indicators of LSTIDs.
To approach the problem of MSTID forecasting, a climatology of the variability of detrended Total Electron Cnotent (TEC) was established (Figure 2), and extreme MSTIDs are considered present when the observed dTEC falls outside the middle quartiles of variability. This method was tested during periods of strong tropospheric disturbance over central Europe, and periods of substorm activity, when MSTIDs were known to be present from the analysis of ionosonde and Doppler sounding data.
The TIDforecasting algorithms and all the software codes are provided with open access.
The efficiency of the T-FORS products are tested with on ground demonstrations. The experiments use one transmitter of the Nostradamus array in France and as receiver the direction-finding (DF) system in Germany.
The efficiency of the receiver was verified based on the azimuth of arrival of the transmitted signals. In most of the cases of failures in receiving the transmitted signal, T-FORS models using the TFT Machine Learning method, forecasted the occurrence of TID activity over Dourbes Digisonde that is located very close to the reflection point between the Transmitter and Receiver in this experiment.
Following this confirmation phase, the final T-FORS forecasting codes are released with Open Access. A roadmap is also proposed for the transition to operations, including a list of recommended High Level Data Products that meet the ESA and WMO requirements.
