Speaker
Description
Geomagnetic indices such as the Dst index are relevant for quantifying global geomagnetic storm impacts, yet their operational forecasting remains constrained by reliance on solar wind measurements from L1-based spacecrafts like ACE or DSCOVR. Current models depend on near-real-time L1 data to make the predictions, creating a fundamental limitation: the magnitude of the geomagnetic storm cannot be forecasted until solar wind disturbances reach ACE. This delay inherently restricts the forecast horizon, as extending predictions beyond the time it takes for storms to propagate from L1 to Earth risks excluding the critical initial phase of the disturbance.
We have evaluated the practical limits of forecasting the Dst index using L1 data. We quantify this constraint by analyzing recent intense storms (May and October 2024) with a neural network trained on ACE plasma and IMF parameters to forecast the Dst index 1 to 6 hours ahead. We observe that predictions initialized before a storm’s arrival at ACE fail to capture the onset and main phase of the storm. Our analysis reveals a reliable forecast horizon of 3 hours, beyond which predictive skill for the main phase degrades sharply. However, longer horizons (>6 hours) remain viable for forecasting the storm recovery phase.
During the May storm of 2024, at 2024-05-10 16:00:00 UTC the disturbance had not yet reached L1 (the first measurement of an active proton speed was at around 16:40). As such the Dst forecasting model predicted little to no variation in the index, since the situation, at that time, was similar to inactive time. However, the Dst will take values of +25, +66, -33, -131, -157 and -277 in the next 6 hours. This presents a critical limitation: a disturbance will start in the upcoming hours, but it has not yet reached ACE, therefeore it is unfeasible to forecast. At 2024-05-10 17:00:00 UTC the first few measurements of the disturbance have already reached L1 and the model starts to properly forecast the main phase of the storm. However, the model requires another hour of data to make a reliable forecast of the disturbance. A similar situation can be observed during the October 2024 storm, where for the longest time horizon the start of the main phase of the storm was not forecasted reliably.
In contrast, the forecasts of the model up to 6 hours ahead show a good accuracy with the Dst index during the recuperation times, as for those changes all the relevant information has already been measured by ACE.
| Do you plan to attend in-person or online? | In-person |
|---|
