Dumbović et al. (Solar Physics 290, 579-612, 2015) derived from a 211-event SOHO-era sample an empirical model that maps five remotely observed CME/flare parameters (plane-of-sky speed, source-region distance from disk center, apparent angular width, associated soft X-ray flare class, and likelihood of CME-CME interaction) to a probability distribution over geomagnetic storm intensity bins of the Dst index, combining parameter-conditioned geometric distributions under an independence assumption and adjusting them with additive constants. The authors state that in this form the model cannot be used for forecasting. The open problem is to complete this remote-observation-only framework into a validated forecasting capability: calibrated probabilistic and thresholded storm-intensity predictions with quantified hit, miss, and false-alarm rates. This jointly requires (i) defining and validating decision thresholds on the probability output, (ii) disentangling CMEs that never reach Earth from those that arrive with unfavorable magnetic-field orientation, and (iii) quantifying the bias induced by positively correlated solar parameters in the independence-based probability combination; the model's development-sample-fitted calibration constants and its unmodeled Sun-Earth propagation and projection effects pose further difficulties.

Background

A coronal mass ejection (CME) is an eruption of magnetized plasma from the solar corona; its interplanetary counterpart (ICME) is the principal driver of the largest geomagnetic storms. Storm intensity is conventionally measured by the Dst index, an hourly index built from low-latitude magnetometer stations that tracks the strength of the magnetospheric ring current in nanotesla (nT); deep negative excursions (commonly ∣Dst∣≥100|Dst| \geq 100 nT) mark intense storms. "Geoeffectiveness" denotes the capacity of a solar eruption to produce a given storm level. The dominant near-Earth driver is the southward component of the ICME magnetic field (negative BzB_z, the north-south component of the interplanetary magnetic field, IMF), which enhances solar-wind-magnetosphere reconnection and injects energy into the ring current. Because the ICME magnetic field cannot be measured before the structure reaches monitoring spacecraft near the Sun-Earth L1 libration point, and L1 measurements themselves give only about one hour of warning, storm-intensity forecasts with one-to-three-day lead time must rest on solar-side ("remote") observations alone.

The Solar and Heliospheric Observatory (SOHO) carries the Large Angle and Spectrometric Coronagraph (LASCO) instruments, whose CME catalog lists for each event a first-order (linear-fit) speed measured in the plane of the sky, a projected quantity subject to projection effects unless the CME erupts near the solar limb, and an apparent angular width, with a 360-degree "halo" width suggesting a launch toward or away from Earth. Complementary disk imagery gives the associated soft X-ray flare class recorded by the Geostationary Operational Environmental Satellites (GOES), graded B, C, M, X on a logarithmic scale and the source-region distance from disk center; successive eruptions can also interact (CME-CME interaction), altering propagation and impact.

Earlier studies had established qualitative links between these remote parameters and storm response, mostly on ICME- or storm-selected samples that exclude the dominant outcome: most CMEs never reach Earth or arrive with an unfavorable magnetic orientation and produce no storm. Dumbović et al. (2015) quantified the relations on a CME-based sample of 211 CME-flare pairs with associated Dst minima: mean storm strength grows roughly linearly with plane-of-sky speed and follows power-law dependences on source-center distance and interaction level. From these they built the model at issue: discrete geometric (exponentially decaying) distributions over four storm-intensity bins (∣Dst∣<100|Dst| < 100, 100-200, 200-300, >300> 300 nT) conditioned on each parameter bin, adjusted by additive constants (their Table 6) to match observed bin frequencies, and combined across parameters through an inclusion-exclusion rule (their Eq. 7) derived by assuming the parameters are independent. The output is an ensemble probability distribution over storm intensity for a new CME/flare parameter set, always skewed toward the non-geoeffective bin. The authors state the model in this form cannot be used for forecasting: probability thresholds and their alarm statistics require further study; non-arriving CMEs and arriving-but-unfavorably-oriented CMEs are indistinguishable in the statistics; the independence assumption (the parameters are known to be positively correlated, e.g., CME speed and flare class) can inflate geoeffectiveness probabilities and false alarms, unquantifiably at present; the additive adjustment constants lack a systematic estimation and stability analysis; and plane-of-sky inputs plus purely time-window-based allowances for drag-based deceleration and flank impacts leave the Sun-Earth propagation unmodeled. Neighboring capabilities bracket the problem but do not fill the gap: drag-based and magnetohydrodynamic propagation models still mispredict CME arrival by roughly 10-15 hours under operational defaults (Yordanova et al. 2024), the ISWAT/COSPAR-oriented review by Temmer et al. (2023) lists reliable prediction of Earth-impacting structures among open questions for the next 5-10 years, machine-learning classifiers fed with in-situ ICME measurements at 1 au classify storm intensity well (Ye et al. 2024) but only after arrival, physics-based couplings of propagation models to magnetospheric response for Dst remain in two-event validation (Maharana et al. 2024), the same group's operational COMESEP alert system (CACTus detection plus drag-based model plus geoeffectiveness tool) was validated in 2017 with an unacceptably low success rate and unacceptably high false-alarm rate (Dumbović et al. 2017), and deep-learning classifiers conditioned on Earth-arriving CMEs (Alobaid et al. 2024) isolate the geoeffectiveness question from the arrival question rather than solving the joint problem.

Problem Statement

The problem, as posed by the source work, is to convert the empirical remote-observation-based probabilistic description of CME geoeffectiveness into a validated forecasting capability. Concretely: construct a procedure that, given only observables available at eruption time (plane-of-sky CME speed and apparent angular width from coronagraphs, source-region position and associated soft X-ray flare class from disk observations, and an assessment of likely CME-CME interaction), issues calibrated predictions of the geomagnetic response at Earth, either as a probability distribution over storm-intensity bins of the Dst index (the hourly low-latitude geomagnetic disturbance index, in nT, with deep negative values marking stronger storms) or as a thresholded storm/no-storm and intensity decision, and demonstrate its validity on events not used in model construction. Meeting this single objective jointly requires: (i) defining probability or decision thresholds on the model output and validating the resulting hit, miss, and false-alarm rates against observed outcomes; (ii) disentangling, within the non-geoeffective class, CMEs that never reach Earth from those that arrive with an unfavorable magnetic-field orientation, since these two failure modes dominate the false-alarm budget and require different physics to predict; and (iii) correcting or replacing the independence assumption used to combine parameter-conditioned probabilities, by quantifying the bias that positively correlated solar parameters (for example CME speed and flare class) induce in the joint probabilities for large storms. The source describes two further limitations of the present model that the objective does not pose as requirements: its additive calibration constants (their Table 6) are fitted to the development sample without a systematic uncertainty or cross-epoch stability analysis, and the Sun-Earth propagation dynamics - drag-based deceleration and flank impacts - together with projection effects in the plane-of-sky input observables are not explicitly modeled, so launch-time measurements are not translated into a dynamical prediction of arrival and near-Earth magnetic structure; the arrival-versus-orientation discrimination required by (ii) is instead to be obtained from the remote observables within this framework. A complete answer establishes, with out-of-sample validation statistics (hit/miss/false-alarm rates, probabilistic calibration and skill relative to a climatological baseline), that such a tool achieves calibrated forecast skill from remote observations alone.

References

  1. M. Dumbović, A. Devos, B. Vršnak, D. Sudar, L. Rodriguez, D. Ruždjak, K. Leer, S. Vennerstrøm, A. Veronig (2015). Geoeffectiveness of Coronal Mass Ejections in the SOHO Era. Solar Physics 290, 579-612. DOI: 10.1007/s11207-014-0613-8, https://doi.org/10.1007/s11207-014-0613-8 (preprint: arXiv:1410.3303, https://arxiv.org/abs/1410.3303)
  2. M. Dumbović, N. Srivastava, Y. K. Rao, B. Vršnak, A. Devos, L. Rodriguez (2017). Validation of the CME Geomagnetic Forecast Alerts Under the COMESEP Alert System. Solar Physics 292, 96. DOI: 10.1007/s11207-017-1120-5, https://doi.org/10.1007/s11207-017-1120-5
  3. M. Dumbović, B. Vršnak, J. Čalogović (2015). Forbush Decrease Prediction Based on Remote Solar Observations. Solar Physics 291, 285-302. DOI: 10.1007/s11207-015-0819-4, https://doi.org/10.1007/s11207-015-0819-4
  4. M. Temmer, C. Scolini, I. G. Richardson, S. G. Heinemann, E. Paouris, A. Vourlidas, M. M. Bisi, N. Al-Haddad, T. Amerstorfer, L. Barnard, D. Burešová, S. J. Hofmeister, K. Iwai, B. V. Jackson, R. Jarolim, L. K. Jian, J. A. Linker, N. Lugaz, P. K. Manoharan, M. L. Mays, W. Mishra, M. J. Owens, E. Palmerio, B. Perri, J. Pomoell, R. F. Pinto, E. Samara, T. Singh, D. Sur, C. Verbeke, A. M. Veronig, B. Zhuang (2023). CME propagation through the heliosphere: Status and future of observations and model development. Advances in Space Research. DOI: 10.1016/j.asr.2023.07.003, https://doi.org/10.1016/j.asr.2023.07.003
  5. E. Yordanova, M. Temmer, M. Dumbović, C. Scolini, E. Paouris, A. L. E. Werner, A. P. Dimmock, L. Sorriso-Valvo (2024). Refined Modeling of Geoeffective Fast Halo CMEs During Solar Cycle 24. Space Weather 22, e2023SW003497. DOI: 10.1029/2023SW003497, https://doi.org/10.1029/2023SW003497
  6. Y. Ye, J. Liu, Y. Hao, J. Cui (2024). Evaluating the Geoeffectiveness of Interplanetary Coronal Mass Ejections: Insights from a Support Vector Machine Approach with SHAP Value Analysis. The Astrophysical Journal 972, 52. DOI: 10.3847/1538-4357/ad61d7, https://doi.org/10.3847/1538-4357/ad61d7
  7. B. Vršnak (2021). Analytical and empirical modelling of the origin and heliospheric propagation of coronal mass ejections, and space weather applications. Journal of Space Weather and Space Climate 11, 34. DOI: 10.1051/SWSC/2021012, https://doi.org/10.1051/SWSC/2021012
  8. A. Maharana, W. D. Cramer, E. Samara, C. Scolini, J. Raeder, S. Poedts (2024). Employing the Coupled EUHFORIA-OpenGGCM Model to Predict CME Geoeffectiveness. Space Weather 22, e2023SW003715. DOI: 10.1029/2023SW003715, https://doi.org/10.1029/2023SW003715
  9. K. A. Alobaid, J. T. L. Wang, H. Wang, J. Jing, Y. Abduallah, Z. Wang, H. Farooki, H. Cavus, V. Yurchyshyn (2024). Prediction of Geoeffective CMEs Using SOHO Images and Deep Learning. Solar Physics 299, 159. DOI: 10.1007/s11207-024-02385-w, https://doi.org/10.1007/s11207-024-02385-w
  10. S. Balan, R. Sarkar, N. Srivastava (2025). Assessing Dst prediction models for forecasting the geoeffectiveness of ICME structures. Journal of Astrophysics and Astronomy 46, 54. DOI: 10.1007/s12036-025-10082-8, https://doi.org/10.1007/s12036-025-10082-8
  11. Z. Yan, J. T. L. Wang, H. Wang, H. Lee, J. Jing, Y. Xu, C. Xu, V. Yurchyshyn (2026). Deep Learning-Enabled Prediction of Geoeffective CMEs Using SOHO and SDO Observations. Solar Physics 301, 73. DOI: 10.1007/s11207-026-02665-7, https://doi.org/10.1007/s11207-026-02665-7
  12. Y. Shi, B. Luo, J. Wang, Y. Chen, S. Liu (2026). A Model for Recommending Historical Similar Events in Forecasting CME Geoeffectiveness. Space Weather 24, e2026SW004998. DOI: 10.1029/2026SW004998, https://doi.org/10.1029/2026SW004998