betaFeedback

WorksPhysics and Astronomy

A comprehensive Sun-as-a-star catalog of interplanetary type-III bursts: Statistics and a framework for stellar exospace weather applications

Abstract

Decameter-hectometric (DH) type-III bursts, produced by energetic electron beams escaping into interplanetary (IP) space, are often associated with strong soft X-ray (SXR) flares, coronal mass ejections (CMEs), and solar energetic particle events (SEPEs). With future DH missions being planned, these intense bursts could provide reliable exospace weather event diagnostics, given the lack of direct stellar IP particle and CME probes. We present a catalog of DH type-III bursts associated with 538 out of 734 ≥\geqC5-class flares observed by Wind and STEREO spacecraft during 2006-2025. Independent of the SXR flare strength, the occurrence and Sun-as-a-star emission properties of DH type-III bursts distinguish confined flares, CMEs with and without strong IP shocks, and SEPEs. We find that logistic-regression models combining SXR and type-III properties provide more accurate and well-calibrated CME/SEPE occurrence probabilities than a model based only on SXR parameters. Further, a space weather event probability-estimation model based on physically meaningful, scale-free parameters derived from SXR and type-III properties outperforms an SXR parameter-only model. This result presents a way to define solar-calibrated exospace weather prediction models extendable to stellar flares with a range of energy scales but similar underlying physics. We also present a catalog of 26 shock-associated type-III (S-type-III) bursts that appear to start from cotemporal DH type-II bursts and are generally associated with fast halo CMEs and major SEPEs. Our catalogs and the scale-free model form a benchmark for exospace weather studies.

Introduction

Type-III bursts are the most common solar radio bursts, appearing as rapidly drifting bright features in dynamic spectra ([82, 53, 68]). They are produced by nonthermal electron beams via coherent plasma emission processes at harmonics of the local plasma frequency ([22, 54, 12] (V. N. Tsytovich & S. A. Kaplan 1969)). Metric type-III bursts trace electron beams propagating in the lower corona, while decameter–hectometric (DH; 1–14 MHz) and kilometric bursts probe particles escaping into interplanetary (IP) space and are therefore particularly relevant to space weather studies ([23]).

Flare-reconnection events often produce isolated or groups of type-III bursts lasting several minutes, referred to as complex type-III bursts ([70, 23, 68]). Type-III storms lasting for ∼\simhours–days form another class of events associated with interchange reconnection near the Corresponding author: Atul Mohan

[email protected] boundaries of active regions and open magnetic field structures [20, 75, 6, 19]. Rarely, a subclass of type-III bursts is found that originates and drifts down from the DH type-II burst features in the dynamic spectrum [69, 35, 13]. We refer to these shock-associated events as S-type-III bursts, which can diagnose shock-accelerated electrons and other energetic particles [e.g. 14, 35, 71]. Figure 1 illustrates the various kinds of

Dynamic spectrum illustrating type-III storm, complex type-III, type-II, and S-type-III bursts

Figure 1. Illustration of the various kinds of DH type-III bursts. A type-III storm is seen before the start of a complex type-III burst associated with an M6.8 class flare. The event produced a halo CME which caused the DH type-II and S-type-III bursts.

DH type-III bursts during a GOES M6.8 class flare associated with a fast (1813 km s−1^{-1}) halo CME and a major solar energetic particle event ( SEPE). A major SEPE is one with >10>10 pfu proton flux in the >10>10 MeV band (1 pfu = 1 particle cm−2^{-2}s−1^{-1}). This study focuses on isolated and complex DH type-III and S-type-III bursts.

Owing to their coherent emission mechanism, type-III bursts associated even with weak flares can outshine the solar or stellar disk emission by several orders of magnitude [e.g. 72, 1, 61], making them excellent Sun-as-a-star diagnostics of energetic electrons and associated space weather events [e.g. 49, 56]. Compared to radio burst types in various spectral bands, DH type-III bursts, particularly complex type-III bursts, are more closely associated with major SEPEs [46, 13, 55] (R. J. MacDowall et al. 2003). However, robust relations between DH type-III burst parameters and the occurrence of CMEs or SEPEs with various properties have not been found. This is likely because major SEPEs are produced by CME-driven shocks, whose direct radio diagnostic is a type-II burst [e.g. 24]. Nevertheless, DH type-II burst events co-temporal with type-III bursts are reported to exhibit particularly high SEPE association rates [17, 83]. Existing studies therefore suggest that combining DH type-III diagnostics with co-temporal radio bursts and soft X-ray (SXR) flare properties may improve predictions of SEPE, CME, and strong IP shock occurrence and properties [26] (R. J. MacDowall et al. 2003).

In stellar studies, where in-situ measurements and active region imaging are unavailable, coherent radio bursts can provide reliable diagnostics of flare reconnection, CMEs, and SEPEs. Type-III bursts, being relatively brighter than type-II bursts, could serve as a sensitive stellar space weather diagnostic, which remains to be tested using solar datasets. Meanwhile, type-II, III, and IV bursts remain largely elusive in active stellar flare monitoring campaigns [e.g. 38, 80], with only recent reports of stellar type-III and type-IV bursts [62]. This dearth has prompted debates about whether radio-quiet major SXR flares, CMEs, and SEPEs could exist on the Sun or whether stellar flares in some active stars follow different emission mechanisms. Meanwhile, solar observations show that the non-detection of a radio burst during a flare does not necessarily mean that the event is intrinsically radio quiet. Propagation effects and line-of-sight absorption can suppress the emission [e.g. 9, 39, 67, 77], while coherent emission is also inherently beamed. Using a long-term multi-spacecraft catalog of DH type-IV bursts, A. Mohan et al. [57] showed that, for any burst event without occulting plasma structures along the lines-of-sight of all simultaneously observing spacecraft, the brightest and most extended event is detected by the spacecraft viewing the active region within ±60∘\pm60^{\circ} longitude in its field of view.

This study aims to build a multi-spacecraft catalog of DH type-III bursts associated with ≥\geC5-class flares using decades-long observations. Separate catalogs of flares without type-III bursts and S-type-III bursts are constructed. We present a statistical analysis of the type-III bursts and explore their relations with flare and space weather events. We further develop probabilistic models to assess whether DH type-III properties improve CME and SEPE occurrence prediction beyond SXR flare properties, and explore a scale-free Sun-as-a-star framework extendable to stellar flares that vary in physical scales but share common physics. Combined with the multi-spacecraft DH type-II (N. Gopalswamy et al. [28]; A. Mohan et al. [60, 59]) and type-IV (A. Mohan et al. [57]) catalogs, this work forms a comprehensive Sun-as-a-star database of IP radio bursts and associated flares, CMEs, and SEPEs for solar and stellar space weather studies. Our catalogs and statistical analysis could guide solar and stellar space weather assessment using the planned DH-band facilities (e.g. M. J. Bentum et al. [8]; M. Knapp et al. [47]; D. R. DeBoer et al. [18]). Section 2 describes the data, while Sect. 3 explains the catalog building and analysis methodology. Section 4 presents the catalog and statistics. Section 5 discusses the implications and explores the importance of DH type-III properties in solar and stellar space weather assessment. Our conclusions are presented in Sect. 6.

Data

The primary data sources for this work are the SXR data in the 0.5 – 8\AA range from GOES, DH dynamic spectra from the Radio and Plasma Wave Investigation (WAVES; J. L. Bougeret et al. [10]) instrument on board Wind, and the Solar TErrestrial RElations Observatory (STEREO; R. A. Howard et al. [42]) spacecraft. Given the availability of STEREO data, we focus on flares that occurred between December 2006 and January 2025. We use SXR flare, CME, and major SEPE catalogs maintained by NASA’s Coordinated Data Analysis Workshops (CDAW) Data Center1.

Using these catalogs, we prepared an initial event list of ≥\geC5-class flares that occurred within ±55∘\pm55^\circ longitude on the solar disk. We impose this longitude restriction to minimize the impact of line-of-sight effects on burst detectability and to enable reliable magnetic field analysis of flare sites. For each flare in our initial list, the basic parameters like GOES class, start time (tstartt_{\mathrm{start}}), peak time (tpeakt_{\mathrm{peak}}), rise time (trise=tpeak−tstartt_{\mathrm{rise}}=t_{\mathrm{peak}}-t_{\mathrm{start}}), and source (active region) location are recorded along with the properties of associated CMEs and SEPEs.

Methodology

DH type-III bursts were searched blindly during all SXR flares in the initial list to enable an unbiased search for radio diagnostics of particle acceleration, CMEs, and major SEPEs during every flare. Events missed by this search strategy are mainly those associated with weak SXR flares (<<C5) or radio bursts with negligible excess SXR emission (e.g. C. E. Alissandrakis et al. [1]; A. Mohan et al. [61]; C. Cattell et al. [15]). Individually, such weak events are minor for space weather, though they may contribute collectively to the steady accelerated-particle background (M. J. Aschwanden & S. L. Freeland [2]), like type-III storms (e.g. J. Fainberg & R. G. Stone [21]; M. G. Aubier et al. [6]; L. Harra et al. [36]). However, we will miss filament eruptions, often associated with weak flares, which can cause SEPE-producing shocks at large IP distances (N. Gopalswamy et al. [27]).

We retrieved data from the GOES database to generate SXR flare light curves in the 0.5–4 and 1–8 Å channels for every event in the initial catalog. Calibrated co-temporal DH dynamic spectra and 1 MHz light curves were generated for all events, using archival data in the Wind/WAVES and STEREO/WAVES repositories managed by NASA’s Space Physics Data Facility (SPDF)2. Python codes were developed to automate these tasks. For each SXR flare, calibrated co-temporal DH dynamic spectrum and the 1 MHz light curve were manually analysed to identify the associated DH type III bursts between tstart−0.5t_{\mathrm{start}}-0.5 h and tpeak+1t_{\mathrm{peak}}+1 h. If the flare subsides to the background level before tpeak+1t_{\mathrm{peak}}+1 h, we limit the search window to the flare’s end time. If no burst is identified within the search window, the event is moved to a separate catalog of flares with no/insignificant DH type-III bursts. If multiple type-III features are present within the search window, the burst that starts within the tstartt_{\mathrm{start}} to tpeak+trise/2t_{\mathrm{peak}}+t_{\mathrm{rise}}/2 and shows a distinctive enhanced flux density is chosen. This is because a strong type-III burst feature is expected from the standard flare model during this period if open magnetic field structures are associated with the reconnection site (e.g. T. Hirayama [40]; K. Shibata et al. [73]; M. J. Aschwanden et al. [3]). However, periods of overlapping type-III bursts or SXR flares, like in Figure 2, make the SXR–radio association ambiguous. We exclude such periods from the analysis. For each identified type-III burst, the start and end times, peak flux density (Fr,pF_{\mathrm{r,p}}), and integrated flux density (Fr,intF_{\mathrm{r,int}}) are computed from the 1 MHz light curve. The type-III peak time relative to the SXR flare peak time (δtpeak\delta t_{\mathrm{peak}}) is recorded. A negative δtpeak\delta t_{\mathrm{peak}} implies that the radio emission peaks before SXR. The properties of CMEs (mean projected speed (VCMEV_{\mathrm{CME}}), width), the flux of SEPEs, and the occurrence of ground-level enhancements (GLEs) associated with each event are obtained from the catalogs maintained by CDAW (S. Yashiro et al. [85]; N. Gopalswamy et al. [32, 29]). If no SEPE is detected, the flux is reported as -1, while a value of 0 denotes enhanced background activity that prevents a reliable SEPE association. For GLEs, we report either 1 to denote an event occurrence or -1 otherwise.

Calibrated DH dynamic spectrum and SXR light curves showing overlapping DH bursts and SXR flares

Figure 2. A period of overlapping DH bursts and SXR flares. Calibrated DH dynamic spectrum and the 1 MHz data show multiple type-III spikes, while the SXR light curve shows signs of overlapping flares. The tstartt_{\mathrm{start}} and tpeak+1t_{\mathrm{peak}}+1 h epochs for the strongest SXR flare during the period are marked by vertical dashed lines, while its tpeakt_{\mathrm{peak}} epoch is marked by a black line.

For every event in the DH type-III catalog, we defined four quality metrics to quantify (i) the confidence in the radio-SXR association (Pk-rank) based on δtpeak\delta t_{\mathrm{peak}}, (ii) the reliability of the radio peak flux identification (M-score), (iii) the radio data quality (data_qlty) based on δtpeak\delta t_{\mathrm{peak}}, and (iv) the confidence in the flare - CME association (Ev_complexity). Pk-rank of 1 to 4 were assigned when the radio peak occurs in the intervals (tstartt_{\mathrm{start}}-0.5 h) to (tstart+trise/2t_{\mathrm{start}}+t_{\mathrm{rise}}/2), (tstart+trise/2t_{\mathrm{start}}+t_{\mathrm{rise}}/2) to (tpeak−Δtt_{\mathrm{peak}}-\Delta t), (tpeak−Δtt_{\mathrm{peak}}-\Delta t), and (tpeak+Δtt_{\mathrm{peak}}+\Delta t) to (tpeak+1t_{\mathrm{peak}}+1 h), respectively, where Δt=min⁡(3 min,trise/2)\Delta t=\min(3\,\mathrm{min},t_{\mathrm{rise}}/2). Larger ranks indicate progressively weaker radio-SXR association. Type-III bursts can be complex with multiple peaks, or there can be overlapping bursts. M-score rates the confidence in the burst identification and estimation of the peak radio flux. When a single burst with a distinguishable peak can be identified, M-score=1. A score of 2 is assigned when reliable radio-SXR event associations can be made, despite the presence of multiple temporally close radio peaks. The peak radio flux density estimates for these events are reliable, but not the integrated flux. A M-score of 3 denotes a complex event for which the radio-SXR event association is uncertain, and the peak and integrated flux density estimates are unreliable. The data_qlty metric quantifies the radio data quality. A value of 0 denotes poor data with substantial gaps that affected peak and integrated flux density estimates; 1 denotes data with gaps but reliable peak flux estimates, while integrated fluxes may be unreliable; and 2 denotes good-quality data suitable for both peak and integrated flux density measurements. Finally, the Ev_complexity takes a value of 0 if the flare and CME identifications are robust, 1 if the CME measurements are uncertain, 2 if the SXR flare profile is impacted by co-temporal emission though FX,pF_{\mathrm{X,p}} and peak time estimates are robust, and 3 if CME uncertainty and flare profile contamination are present.

The relationships between DH type-III burst properties and associated space weather events were examined using events with reliable measurements. Additionally, we obtained the DH band data from STEREO A and B spacecraft for every event in the DH type-III catalog and in the catalog of flares without type-III bursts and generated the calibrated dynamic spectra and 1 MHz light curves. The data from the period when the STEREO spacecraft were significantly apart from Wind offer an opportunity to study the impact of line-of-sight on the observed radio bursts.

To compile a list of S-type-III bursts, we visually inspected all events in the DH type-II catalog3 (N. Gopalswamy et al. [28]), maintained by CDAW, starting from late 1997 to 2025-01-01. We did not apply any selection criteria, including in the source longitude range, to the DH type-II catalog because of the relatively low event count compared to SXR flares. We assign a ‘confidence’ metric to the detected S-type-III events, where 0 denotes a less confident detection, and 1 denotes a confident detection. The CME and flare characteristics, along with the peak flux of any associated SEPE, are tabulated following the same procedure used for the DH type-III catalog.

Results

We identified 901 ≥\geqC5-class flares with co-temporal SXR and DH band data from the GOES and Wind spacecraft, respectively. After removing events that are either complex or have poor-quality data, we obtain a list of 734 events, of which 311 (42%) are eruptive (CME-associated), and 36 CMEs are major SEPE-associated. Of the 734 flares, 538 (73%) have associated DH type-III bursts, constituting our DH type-III catalog. We present an additional catalog of 196 flares without robust type-III emissions, of which 180 (88%) are confined. Meanwhile, among the 675 DH type-II bursts in the CDAW catalog, we found 26 (4%) events with S-type-III bursts. We provide a catalog of these events with the properties of the associated CMEs, SXR flares, and SEPEs. The catalogs are supplemented by calibrated 1 MHz and SXR light curves, DH dynamic spectra (from the Wind, STEREO A, and STEREO B spacecraft), SEPE flux light curves, CME height-time plots, and links to white light and SXR movies of the Sun. All catalogs and associated data are available online4. Appendix § 9 presents screenshots of all catalogs.

DH type-III catalog

Table 1 presents the properties of SXR flares, radio bursts, and associated space weather events, reported in the catalog. Additionally, the catalog provides the start and end times of the type-III bursts and SXR flares. Figure 3 shows the histograms and respective median values of the various type-III burst parameters.

PropertyDescription
Type-IIIDur_min (ΔTburst\Delta T_{\mathrm{burst}})Duration in min
PeakFlux_SFU (Fr,pF_{\mathrm{r,p}})Peak flux in SFU (Catalog mentions log⁡10Fr,p\log_{10}F_{\mathrm{r,p}})
IntegFlux_SFUs (Fr,intF_{\mathrm{r,int}})Integrated flux in SFU s
Time-wrt-pflr_min (δtpeak\delta t_{\mathrm{peak}})Radio burst peak time relative to the SXR flare peak
SXR flaret_rise_min (triset_{\mathrm{rise}})Flare rise time in min
ClassGOES flare class
FXray (FX,pF_{\mathrm{X,p}})Peak SXR flux in W m−2^{-2}
CMEVcme (VCMEV_{\mathrm{CME}})Projected CME speed in km s−1^{-1}
WidthCME angular width in degree
DHtypeIIOccurrence of DH type II burst, where 1 means yes and 0 means no
SEPESEP_maxPeak SEPE flux in pfu
GLEGLEGLE occurrence, where 1 means yes and −1 means no

Table 1. Properties of type-III bursts, flares, CMEs, SEPEs, and GLE events in the DH type-III catalog.

Histograms of properties of all DH type-III bursts in the catalog

Figure 3. Properties of all DH type-III bursts in the catalog. Data count and median values are also provided.

Figure 4 presents the properties of flares, CMEs, and SEPEs associated with the events in the DH type-III catalog. The median flare class is M1.4 and the median SEPE peak flux is 58 pfu. The 2-dimensional (2D) histogram of

Figure 4. Properties of all events in the DH type-III catalog. CMEs with reliable measurements were used to make the 2D histogram in panel (b).

CME characteristics reveals two distinct populations in the VCMEV_{\mathrm{CME}}–width plane with different median properties: (1) slow-narrow CMEs and (2) fast-halo CMEs. To understand the two populations, Figure 5 plots the 2D histogram of

Two-panel 2D histograms showing the joint distribution of CME width and CME speed

Figure 5. 2D histograms showing the joint distribution of the width and VCMEV_{\mathrm{CME}} of CMEs with and without DH type-II bursts in the DH type-III catalog.

VCMEV_{\mathrm{CME}} and width for CMEs associated with IP shocks or DH type-II bursts and those without. Clearly, the fast-halo population is dominated by CMEs associated with DH type-II bursts, while the remaining events form a largely slow, non-halo population, consistent with previous studies [31, 34]. For a given coronagraph, halo CMEs are known to be relatively faster with stronger SXR flares than a regular CME [30]. This is because only energetic CMEs expanding fast enough to appear to surround the occulting disk are observed as halos. However, the direction of CME propagation relative to the observer is a major factor in making a fast CME observable as a halo. In our sample, only CMEs originating from active regions within ±55∘\pm55^\circ longitude are included, excluding events far from the disk center. Hence, VCMEV_{\mathrm{CME}} may play a greater role in making a CME observable as a halo. A detailed case-by-case analysis, beyond the scope of this study, would be required to explore this possibility further. We therefore use the occurrence of CME shock-driven DH type-II bursts as a diagnostic of energetic fast-CMEs, rather than a heuristic based on the VCMEV_{\mathrm{CME}}–width plane.

Figure 6 explores the DH type-III properties within four space weather relevant event groups, namely: ‘confined’: confined flares, ‘CME-no-DHII’: CMEs without DH type-II bursts, ‘CME+DHII’: CMEs with DH type-II bursts, and ‘SEP’: SEPEs. To retain only reliable measurements, the rest of our analysis uses bursts with MM-score=1 and starting before the SXR flare tpeakt_{\mathrm{peak}}. Although the association of confined flares with type-III bursts may appear surprising, these bursts can arise from jets or minor reconnections involving nearby magnetic structures, while most flare-accelerated particles are confined to the active region (see, [16] (T. Li et al. 2019)). Histograms of all burst properties except δtpeak\delta t_{\mathrm{peak}} show signs of separation between event groups, notably in their lower envelopes. The bursts associated with confined flares show the lowest median Fr,intF_{\mathrm{r,int}} and ΔTburst\Delta T_{\mathrm{burst}}, indicating that most of the energy in the flare-accelerated electron beams is dissipated close to the flare site. The median values of ΔTburst\Delta T_{\mathrm{burst}}, Fr,pF_{\mathrm{r,p}}, and Fr,intF_{\mathrm{r,int}} increase systematically from confined to CME-no-DHII, CME+DHII, and SEP groups (Table 2). These inferences from the histogram analysis, together with the finding that 88% of the SXR flares without associated DH type-III emission are confined, suggest that both the occurrence and properties of flare-associated DH type-III bursts may carry useful space weather diagnostic information.

PropertyConfinedCME-no-DHIICME+DHIISEP
ΔTburst\Delta T_{\mathrm{burst}} [min]12.021.037.551.0
Fr,pF_{\mathrm{r,p}} [SFU]6.3×1046.3 \times10^{4}3.5×1053.5 \times10^{5}1.3×1061.3 \times10^{6}6.3×1066.3 \times10^{6}
Fr,intF_{\mathrm{r,int}} [SFU s]1.4×1071.4 \times10^{7}5.8×1075.8 \times10^{7}3.3×1083.3 \times10^{8}1.2×1091.2 \times10^{9}
δtpeak\delta t_{\mathrm{peak}} [min]−1.3-1.3−0.2-0.2−2.8-2.8−3.0-3.0

Table 2. Median values (computed in linear space) of the various DH type-III properties for the different event groups.

Figure 6. Histograms of various DH type-III properties for the four different event groups. Distributions for all properties, except δtpeak\delta t_{\mathrm{peak}}, show a significant separation in their lower envelope and median value across groups.

Discussion

This section explores whether the distributions of DH type-III burst properties are statistically distinct across the four event groups and whether these properties, together with flare parameters, can be used to develop statistical models to estimate the occurrence probabilities of major space weather events. Meanwhile, models based on extrinsic quantities like Fr,intF_{\mathrm{r,int}} and ΔTburst\Delta T_{\mathrm{burst}}, which depend on Sun-specific physical scales, cannot be generalized to stellar systems. So, we will discuss ways to define scale-free Sun-as-a-star parameters and associated statistical models applicable to exospace weather studies in Sun-like stars. We further discuss the properties of S-type-III bursts and a small fraction of eruptive flares without robust DH type-III association in our flare sample.

Properties of DH type-III bursts

The histogram analysis in Figure 6 suggests that the DH type-III properties Fr,pF_{\mathrm{r,p}}, Fr,intF_{\mathrm{r,int}}, and ΔTburst\Delta T_{\mathrm{burst}} could distinguish the four event groups. If so, these burst properties could improve probabilistic models of space weather event occurrence during ≥\geqC5-class flares. However, we should first analyse the role of the big flare syndrome ([45]) in the results. This is because events with brighter SXR flares typically cause stronger magnetic energy release in the corona and IP space. Consequently, energetic CMEs, stronger IP shocks, and SEPEs tend to be associated with stronger SXR flares. If such flares also preferentially produce brighter and longer-duration DH type-III bursts, then the median properties of type-III bursts can appear to increase going across event groups with weaker to stronger magnetic energy release and particle acceleration. Hence, we should test whether the DH type-III burst properties remain statistically distinct across the different event groups even within narrow intervals of SXR flare strength.

Assessing the impact of the big-flare syndrome

To assess the impact of the big-flare syndrome, we selected the most reliably detected DH type-III bursts (MM-score=1) and divided them into five narrow FX,pF_{\mathrm{X,p}} bins, each containing ≃\simeq100 flares. The log⁡FX,p\log F_{\mathrm{X,p}} ranges of the bins are: (-5.302, -5.097], (-5.097, -4.959], (-4.959, -4.745], (-4.745, -4.387], and (-4.387, -2.824]. However, examining the data counts in individual event groups, we find that CME+DHII and SEP groups contain <5 events in the log FX,p<−4.959F_{\mathrm{X,p}}<-4.959 (<M1.1) bins. Therefore, to improve statistics, all <M1.1 flares in these two groups were merged into a single log FX,pF_{\mathrm{X,p}} bin ((-5.302, -4.959]), while the original FX,pF_{\mathrm{X,p}} bins were retained for stronger flares. Table 5 in Appendix B provides the median properties of DH type-III bursts for various event groups within each FX,pF_{\mathrm{X,p}} bin. Figure 7 shows the median DH type-III burst properties for each event group within every log FX,pF_{\mathrm{X,p}} bin. Error bars show the corresponding median absolute deviations (MADs). The X coordinates indicate the median log FX,pF_{\mathrm{X,p}} values. Though the MAD intervals overlap, the systematic separation of medians of the type-III burst parameters is evident within many FX,pF_{\mathrm{X,p}} bins. This separation is very evident in ΔTburst\Delta T_{\mathrm{burst}}, while Fr,intF_{\mathrm{r,int}} and Fr,pF_{\mathrm{r,p}} can also distinguish the event groups for >M1.1 flares. SEPEs consistently produce the longest-duration DH type-III bursts with the highest peak and integrated flux densities within a given FX,pF_{\mathrm{X,p}} bin. Nevertheless, the apparent tendency of median radio flux to increase with FX,pF_{\mathrm{X,p}} within each group could be due to big-flare syndrome.

Figure 7. Median ΔTburst\Delta T_{\mathrm{burst}}, Fr,pF_{\mathrm{r,p}}, and Fr,intF_{\mathrm{r,int}} values within FX,pF_{\mathrm{X,p}} bins for the different event groups. Error bars indicate the respective MADs. The median type-III burst properties, particularly ΔTburst\Delta T_{\mathrm{burst}}, show systematic differences among the event groups within most FX,pF_{\mathrm{X,p}} bins.

log⁡(FX,p)\log(F_{\mathrm{X,p}}) binPropertyConfinedCME-no-DHIICME+DHIISEP
(−5.302,−5.097](-5.302,-5.097]ΔTburst\Delta T_{\mathrm{burst}} [min]N=72N = 72, Med= 10.0N=19N = 19, Med= 17.0N=3N = 3, Med= 24.0N=0N = 0
Fr,pF_{\mathrm{r,p}} [SFU]N=72N = 72, Med= 2.0×1042.0 \times10^{4}N=19N = 19, Med= 2.8×1052.8 \times10^{5}N=3N = 3, Med= 5.6×1055.6 \times10^{5}N=0N = 0
Fr,intF_{\mathrm{r,int}} [SFU s]N=67N = 67, Med= 4.3×1064.3 \times10^{6}N=18N = 18, Med= 6.4×1076.4 \times10^{7}N=3N = 3, Med= 1.7×1081.7 \times10^{8}N=0N = 0
(−5.097,−4.959](-5.097,-4.959]ΔTburst\Delta T_{\mathrm{burst}} [min]N=56N = 56, Med= 12.5N=36N = 36, Med= 21.0N=3N = 3, Med= 32.0N=0N = 0
Fr,pF_{\mathrm{r,p}} [SFU]N=56N = 56, Med= 1.3×1051.3 \times10^{5}N=36N = 36, Med= 5.5×1055.5 \times10^{5}N=3N = 3, Med= 2.5×1052.5 \times10^{5}N=0N = 0
Fr,intF_{\mathrm{r,int}} [SFU s]N=55N = 55, Med= 1.9×1071.9 \times10^{7}N=36N = 36, Med= 7.0×1077.0 \times10^{7}N=3N = 3, Med= 5.6×1075.6 \times10^{7}N=0N = 0
(−5.302,−4.959](-5.302,-4.959]ΔTburst\Delta T_{\mathrm{burst}} [min]— —N=6N = 6, Med= 28N=0N = 0
Fr,pF_{\mathrm{r,p}} [SFU]— —N=6N = 6, Med= 3.0×1053.0 \times10^{5}N=0N = 0
Fr,intF_{\mathrm{r,int}} [SFU s]— —N=6N = 6, Med= 6.6×1076.6 \times10^{7}N=0N = 0
(−4.959,−4.745](-4.959,-4.745]ΔTburst\Delta T_{\mathrm{burst}} [min]N=38N = 38, Med= 14.0N=42N = 42, Med= 20.5N=10N = 10, Med= 44.5N=3N = 3, Med= 58.0
Fr,pF_{\mathrm{r,p}} [SFU]N=38N = 38, Med= 1.0×1051.0 \times10^{5}N=42N = 42, Med= 3.7×1053.7 \times10^{5}N=10N = 10, Med= 1.3×1061.3 \times10^{6}N=3N = 3, Med= 2.8×1062.8 \times10^{6}
Fr,intF_{\mathrm{r,int}} [SFU s]N=37N = 37, Med= 1.9×1071.9 \times10^{7}N=42N = 42, Med= 7.9×1077.9 \times10^{7}N=10N = 10, Med= 4.4×1084.4 \times10^{8}N=3N = 3, Med= 7.4×1087.4 \times10^{8}
(−4.745,−4.387](-4.745,-4.387]ΔTburst\Delta T_{\mathrm{burst}} [min]N=26N = 26, Med= 14.5N=42N = 42, Med= 22.5N=19N = 19, Med= 34.0N=4N = 4, Med= 44.5
Fr,pF_{\mathrm{r,p}} [SFU]N=26N = 26, Med= 1.0×1051.0 \times10^{5}N=42N = 42, Med= 1.6×1051.6 \times10^{5}N=19N = 19, Med= 1.0×1061.0 \times10^{6}N=4N = 4, Med= 5.8×1065.8 \times10^{6}
Fr,intF_{\mathrm{r,int}} [SFU s]N=26N = 26, Med= 1.6×1071.6 \times10^{7}N=42N = 42, Med= 3.5×1073.5 \times10^{7}N=19N = 19, Med= 1.5×1081.5 \times10^{8}N=4N = 4, Med= 9.6×1089.6 \times10^{8}
(−4.387,−2.824](-4.387,-2.824]ΔTburst\Delta T_{\mathrm{burst}} [min]N=10N = 10, Med= 20.0N=40N = 40, Med= 20.5N=41N = 41, Med= 45.0N=26N = 26, Med= 50.5
Fr,pF_{\mathrm{r,p}} [SFU]N=10N = 10, Med= 1.0×1051.0 \times10^{5}N=40N = 40, Med= 3.5×1053.5 \times10^{5}N=41N = 41, Med= 2.1×1062.1 \times10^{6}N=26N = 26, Med= 1.2×1071.2 \times10^{7}
Fr,intF_{\mathrm{r,int}} [SFU s]N=9N = 9, Med= 1.9×1071.9 \times10^{7}N=40N = 40, Med= 6.6×1076.6 \times10^{7}N=41N = 41, Med= 4.7×1084.7 \times10^{8}N=26N = 26, Med= 2.1×1092.1 \times10^{9}

Table 5. Median type-III burst properties in log FX,pF_{\mathrm{X,p}} bins. Each cell lists the number of events and the median value of each type-III property for a given event group and an FX,pF_{\mathrm{X,p}} bin. The bin ranges in bold highlight the statistics for >M1.1 class flares. Note that the (−5.302,−4.959](-5.302,-4.959] bin is used only for the CME+DHII group to have >5 data points for median computation.

The systematic median separation tells us that the central tendency of a property distribution, or the representative property value, shifts between event groups, for a fixed FX,pF_{\mathrm{X,p}} bin. To assess whether the type-III burst property distributions themselves are statistically distinct across event groups within each FX,pF_{\mathrm{X,p}} bin, we perform Kolmogorov-Smirnov (K-S) tests using the scipy.stats.ks_2samp function in SciPy (P. Virtanen et al. [81]). We consider the following event pairs: (confined, CME-no-DHII), (confined, CME+DHII), (CME+DHII, CME-no-DHII), (SEP, CME-no-DHII). For each type-III property, K-S tests were performed for the selected event pairs within individual log FX,pF_{\mathrm{X,p}} bins. The analysis is restricted to >M1.1 flare bins, where all event groups contain a sizable number of events. Due to the lack of at least five SEPs in the weaker FX,pF_{\mathrm{X,p}} bins, the K-S test for the (SEP, CME-no-DHII) pair was carried out only in the strongest FX,pF_{\mathrm{X,p}} bin.

Table 3 shows the results from the K-S test for each type-III property for every log FX,pF_{\mathrm{X,p}} bin and event pair. Along with the K-S D and p values, the critical D parameters are also found for 95 and 90% confidence levels (Dcrit90D_{\mathrm{crit}}^{90} and Dcrit95D_{\mathrm{crit}}^{95}). For ΔTburst\Delta T_{\mathrm{burst}}, D≥Dcrit95D \ge D_{\mathrm{crit}}^{95} for all event pairs except (confined, CME-no-DHII) within every log FX,pF_{\mathrm{X,p}} bin, clearly revealing that the cumulative distribution functions differ significantly across event groups. The same is true for Fr,intF_{\mathrm{r,int}} and Fr,pF_{\mathrm{r,p}} as well, except for the (CME+DHII, CME-no-DHII) pair in the weakest flare bin. For the (confined, CME-no-DHII) pair, though the distinction between distribution functions is typically moderate to weak for various properties within log FX,pF_{\mathrm{X,p}} bins, we find distinctions with 95% confidence in some bins, especially the weakest flare bin for all parameters and the strongest bins for Fr,intF_{\mathrm{r,int}} and Fr,pF_{\mathrm{r,p}}. These results and the observation that most confined flares lack DH type-III bursts (see Sect. 4) strongly suggest that the Sun-as-a-star type-III emission properties can help significantly improve the estimation of the occurrence probabilities of major space weather events during a solar/stellar SXR flare.

Propertylog⁡(FX,p)\log(F_{\mathrm{X,p}}) binGroup 1N1N_1Group 2N2N_2DDDcrit95D_{\mathrm{crit}}^{95}Dcrit90D_{\mathrm{crit}}^{90}pp
ΔTburst\Delta T_{\mathrm{burst}}(−4.959,−4.745](-4.959,-4.745]Confined38CME-no-DHII420.330.300.272.0×10−22.0 \times10^{-2}
(−4.959,−4.745](-4.959,-4.745]Confined38CME+DHII100.740.480.438.0×10−58.0 \times10^{-5}
(−4.959,−4.745](-4.959,-4.745]CME+DHII10CME-no-DHII420.600.480.433.0×10−33.0 \times10^{-3}
(−4.745,−4.387](-4.745,-4.387]Confined26CME-no-DHII420.350.340.303.0×10−23.0 \times10^{-2}
(−4.745,−4.387](-4.745,-4.387]Confined26CME+DHII190.540.410.372.0×10−32.0 \times10^{-3}
(−4.745,−4.387](-4.745,-4.387]CME+DHII19CME-no-DHII420.420.380.341.0×10−21.0 \times10^{-2}
(−4.387,−2.824](-4.387,-2.824]Confined10CME-no-DHII400.250.480.436.6×10−16.6 \times10^{-1}
(−4.387,−2.824](-4.387,-2.824]Confined10CME+DHII410.730.480.431.1×10−41.1 \times10^{-4}
(−4.387,−2.824](-4.387,-2.824]CME+DHII41CME-no-DHII400.560.300.271.2×10−61.2 \times10^{-6}
(−4.387,−2.824](-4.387,-2.824]SEP26CME-no-DHII400.680.340.311.7×10−71.7 \times10^{-7}
Fr,intF_{\mathrm{r,int}}(−4.959,−4.745](-4.959,-4.745]Confined37CME-no-DHII420.320.310.282.0×10−22.0 \times10^{-2}
(−4.959,−4.745](-4.959,-4.745]Confined37CME+DHII100.530.480.431.0×10−21.0 \times10^{-2}
(−4.959,−4.745](-4.959,-4.745]CME+DHII10CME-no-DHII420.310.480.433.2×10−13.2 \times10^{-1}
(−4.745,−4.387](-4.745,-4.387]Confined26CME-no-DHII420.180.340.306.0×10−16.0 \times10^{-1}
(−4.745,−4.387](-4.745,-4.387]Confined26CME+DHII190.550.410.371.0×10−31.0 \times10^{-3}
(−4.745,−4.387](-4.745,-4.387]CME+DHII19CME-no-DHII420.410.380.341.0×10−21.0 \times10^{-2}
(−4.387,−2.824](-4.387,-2.824]Confined9CME-no-DHII400.420.500.451.1×10−11.1 \times10^{-1}
(−4.387,−2.824](-4.387,-2.824]Confined9CME+DHII410.680.500.458.3×10−48.3 \times10^{-4}
(−4.387,−2.824](-4.387,-2.824]CME+DHII41CME-no-DHII400.380.300.274.0×10−34.0 \times10^{-3}
(−4.387,−2.824](-4.387,-2.824]SEP26CME-no-DHII400.540.340.319.5×10−59.5 \times10^{-5}
Fr,pF_{\mathrm{r,p}}(−4.959,−4.745](-4.959,-4.745]Confined38CME-no-DHII420.320.300.272.0×10−22.0 \times10^{-2}
(−4.959,−4.745](-4.959,-4.745]Confined38CME+DHII100.510.480.432.0×10−22.0 \times10^{-2}
(−4.959,−4.745](-4.959,-4.745]CME+DHII10CME-no-DHII420.280.480.434.9×10−14.9 \times10^{-1}
(−4.745,−4.387](-4.745,-4.387]Confined26CME-no-DHII420.100.340.309.8×10−19.8 \times10^{-1}
(−4.745,−4.387](-4.745,-4.387]Confined26CME+DHII190.480.410.377.6×10−37.6 \times10^{-3}
(−4.745,−4.387](-4.745,-4.387]CME+DHII19CME-no-DHII420.410.380.341.7×10−21.7 \times10^{-2}
(−4.387,−2.824](-4.387,-2.824]Confined10CME-no-DHII400.430.480.439.0×10−29.0 \times10^{-2}
(−4.387,−2.824](-4.387,-2.824]Confined10CME+DHII410.630.480.431.8×10−31.8 \times10^{-3}
(−4.387,−2.824](-4.387,-2.824]CME+DHII41CME-no-DHII400.330.300.271.9×10−21.9 \times10^{-2}
(−4.387,−2.824](-4.387,-2.824]SEP26CME-no-DHII400.450.340.312.2×10−32.2 \times10^{-3}

Table 3. K-S test results (D and p values) for DH type-III properties in log⁡FX,p\log F_{\mathrm{X,p}} bins. Critical D values, DcritD_{\mathrm{crit}}, are computed for 95% and 90% confidence levels using the data counts, N1N_1 and N2N_2.

Space weather event occurrence probabilities from Sun-as-a-star flare diagnostics

Sections 4 and 5.1 show that the occurrence and Sun-as-a-star parameters of DH type-III bursts could help improve the assessment of the space weather impacts of SXR flares. Here, using a sample of 615 SXR flares with reliable detection or non-detection of DH type-III bursts, we assess whether a multidimensional flare vector, f\mathbf{f}, formed using SXR and DH type-III parameters can be used to reliably estimate the occurrence probabilities of different space weather event classes: confined flares, CMEs, and SEPEs. The components of f\mathbf{f} are functions of Fr,pF_{\mathrm{r,p}}, Fr,intF_{\mathrm{r,int}}, ΔTburst\Delta T_{\mathrm{burst}},

triset_{\mathrm{rise}}, FX,pF_{\mathrm{X,p}}, and gg, where gg is a Boolean gate parameter that takes a value of 1 (0) for flares with (without) DH type-III emission. The components of f\mathbf{f} that depend on type-III properties are multiplied by gg to form gated parameters. This gating permits a consistent definition of f\mathbf{f} for flares with and without type-III emission alike. For a space weather event class SiS_i and flare vector f\mathbf{f}, the occurrence probability, P(Si∣f)P(S_i \mid\mathbf{f}), is defined as,

P(Si∣f)=exp⁡(βi,0+βiTf)∑jexp⁡(βj,0+βjTf),(1)P(S_i \mid\mathbf{f}) = \frac{\exp(\beta_{i,0}+\boldsymbol{\beta}_i^{\mathrm{T}}\mathbf{f})}{\sum_j \exp(\beta_{j,0}+\boldsymbol{\beta}_j^{\mathrm{T}}\mathbf{f})}, \tag*{(1)}

where βi,0\beta_{i,0} and βi\boldsymbol{\beta}_i are model coefficients. We consider two probability-estimation problems, Si∈{confined,CME}S_i \in\{\mathrm{confined},\mathrm{CME}\} and Si∈{confined,CME,SEP}S_i \in\{\mathrm{confined},\mathrm{CME},\mathrm{SEP}\}, and various types of probability models that differ in the components of f\mathbf{f} (see Table 4). Some models use purely type-III or SXR flare properties, while others use both or physically-motivated scale-free parameters derived from them to define f\mathbf{f} and infer P(Si∣f)P(S_i \mid\mathbf{f}). For a chosen model type, every flare in the dataset produces an (f,Sobs)(\mathbf{f}, S^{\mathrm{obs}}) tuple, where SobsS^{\mathrm{obs}} is the observed space weather class given the problem. The primary aim of the following analysis is to identify the model type and optimal parameters that provide the most reliable estimates of P(Si∣f)P(S_i \mid\mathbf{f}) for the considered problems.

ProblemModel typeffNNBSCEBA
Confined vs. CMESXR only{FX,p, trise}\{F_{X,p},\,t_{\mathrm{rise}}\}6150.16±0.0010.16 \pm0.0010.04±0.0040.04 \pm0.0040.73±0.0030.73 \pm0.003
Type-III only{g, gFr,p, gFr,int, gΔTburst}\{g,\,gF_{r,p},\,gF_{r,\mathrm{int}},\,g\Delta T_{\mathrm{burst}}\}6150.15±0.0010.15 \pm0.0010.04±0.0060.04 \pm0.0060.77±0.0030.77 \pm0.003
SXR + Type-III{FX,p, trise, g, gFr,p, gFr,int, gΔTburst}\{F_{X,p},\,t_{\mathrm{rise}},\,g,\,gF_{r,p},\,gF_{r,\mathrm{int}},\,g\Delta T_{\mathrm{burst}}\}6150.13±0.0010.13 \pm0.0010.03±0.0040.03 \pm0.0040.80±0.0050.80 \pm0.005
Scale-free{g, glog⁡Fr,pFX,p, glog⁡ΔTbursttrise, log⁡RX}\{g,\,g\log\frac{F_{r,p}}{F_{X,p}},\,g\log\frac{\Delta T_{\mathrm{burst}}}{t_{\mathrm{rise}}},\,\log R_X\}6150.15±0.0010.15 \pm0.0010.03±0.0040.03 \pm0.0040.76±0.0050.76 \pm0.005
Confined vs. CME vs. SEPESXR only{FX,p, trise}\{F_{X,p},\,t_{\mathrm{rise}}\}5870.37±0.0020.37 \pm0.0020.04±0.0030.04 \pm0.0030.56±0.0090.56 \pm0.009
Type-III only{g, gFr,p, gFr,int, ΔTburst}\{g,\,gF_{r,p},\,gF_{r,\mathrm{int}},\,\Delta T_{\mathrm{burst}}\}5870.34±0.0020.34 \pm0.0020.02±0.0040.02 \pm0.0040.67±0.0080.67 \pm0.008
SXR + Type-III{FX,p, trise, g, gFr,p, gFr,int, gΔTburst}\{F_{X,p},\,t_{\mathrm{rise}},\,g,\,gF_{r,p},\,gF_{r,\mathrm{int}},\,g\Delta T_{\mathrm{burst}}\}5870.30±0.0030.30 \pm0.0030.02±0.0030.02 \pm0.0030.71±0.0150.71 \pm0.015
Scale-free{g, glog⁡Fr,pFX,p, glog⁡ΔTbursttrise, log⁡RX}\{g,\,g\log\frac{F_{r,p}}{F_{X,p}},\,g\log\frac{\Delta T_{\mathrm{burst}}}{t_{\mathrm{rise}}},\,\log R_X\}5870.35±0.0020.35 \pm0.0020.02±0.0030.02 \pm0.0030.60±0.0080.60 \pm0.008

Table 4. Performance metrics of various models for the two occurrence probability estimation problems. BS and CE evaluate the occurrence probabilities directly, while BA evaluates the corresponding event predictor based on the maximum-probability class (see Sect. 5.2 for details). Lower values of BS and CE, and higher BA indicate better performance.

For a given problem and a model type, we use stratified five-fold cross-validation with 20 repetitions, given the relatively small sample size, especially in the SEPE class (e.g. [11, 76, 4]). In each repetition, we divide all f\mathbf{f} vectors into five folds that approximately preserve the class fractions in the full sample. Four folds are used to optimize the model coefficients using an L2-regularized logistic regression algorithm ([41]), and the optimized model is then used to estimate P(Si∣f)P(S_i \mid\mathbf{f}) for the flares in the unused test-fold. This is repeated with each of the five folds serving once as the test-fold, yielding for each f\mathbf{f} a set of P(Si∣f)P(S_i \mid\mathbf{f}), i=1,…,Ki = 1, \ldots, K, where KK is the number of space weather classes in the problem. These predicted probabilities and SnobsS_n^{\mathrm{obs}} for every f\mathbf{f} are used to compute different performance scores, detailed in Sect. 5.2.1. This five-fold procedure is repeated 20 times after reshuffling the dataset each time. Each repetition gives one value per performance score, generating a set of 20 values for each score. We report the mean and standard deviation of each score. The whole procedure produces 100 (20×520 \times5) sets of model coefficients, from which we derive an ‘mean-coefficient model’. The entire analysis is repeated for every model type and for both probability-estimation problems, resulting in a mean-coefficient model and a set of three mean performance scores with respective estimation errors.

Performance scores and reliable models for solar and stellar application

The performance scores are evaluated after each five-fold procedure for a probability-estimation problem with KK classes and an overall sample of NN flare vectors. Because the primary objective is to estimate occurrence probabilities rather than assign discrete space weather class labels, we evaluate probability accuracy using the Brier score (BS),

BS=1N∑n=1N∑i=1K[P(Si∣fn)−δSi,Snobs]2(2)\mathrm{BS} = \frac{1}{N} \sum_{n=1}^{N} \sum_{i=1}^{K} \left[P(S_i \mid\mathbf{f}_n) - \delta_{S_i,S_n^{\mathrm{obs}}}\right]^2 \tag*{(2)}

where δ\delta-function is 1 if the iith space weather class, Si=SnobsS_i = S_n^{\mathrm{obs}} and zero otherwise. SnobsS_n^{\mathrm{obs}} denotes the observed class of the nnth flare represented by fn\mathbf{f}_n. BS score hence compares a probability model to an ideal probability-estimator case. Smaller BS indicates more accurate probability estimates. However, BS = 0 is not expected for a stochastic space weather event occurrence problem using Sun-as-a-star parameters that lack spatial information. Further, we assess the reliability of the estimated event probabilities for each SiS_i class using the calibration error (CE) score. Firstly, for the iith event class, we divide the full set of flare vectors into equal-count bins along the axis of predicted probability values and estimate the absolute difference between the observed event occurrence fraction (Op,iO_{p,i}) and the mean probability estimate (Pˉp,i\bar{P}_{p,i}) for each bin, labelled by pp. Combining all such estimates of absolute deviation across bins for every event class, we estimate

CE=1K∑i=1K∑p=1PNpN∣Pˉp,i−Op,i∣(3)\mathrm{CE} = \frac{1}{K} \sum_{i=1}^{K} \sum_{p=1}^{P} \frac{N_p}{N} \left|\bar{P}_{p,i} - O_{p,i}\right| \tag*{(3)}

where Np/NN_p/N measures the fraction of flares in the ppth bin. The CE score estimates the mean deviation of the probability estimates from the observed event occurrence fractions across various event classes considered in a problem. Lower CE indicates better calibration.

As a secondary test, we report the balanced accuracy (BA) score of the class-predictor corresponding to the probability estimation model. The predicted class, S^\hat{S} for a given f\mathbf{f}, is defined as the class with the highest P(Si∣f)P(S_i \mid\mathbf{f}). BA is the mean recall across the KK event classes,

BA=1K∑i=1KTPiTPi+FNi,(4)\mathrm{BA} = \frac{1}{K} \sum_{i=1}^{K} \frac{\mathrm{TP}_i}{\mathrm{TP}_i + \mathrm{FN}_i}, \tag*{(4)}

where TPi\mathrm{TP}_i and FNi\mathrm{FN}_i denote the numbers of true positives and false negatives, respectively, for class SiS_i. BA is used as a common measure of classification accuracy, but it is secondary here because assigning a single class discards the magnitude and distribution of the inferred probabilities.

Solar space weather assessment

For the confined/CME problem, we use all 615 events with the most reliable SXR and radio data: 379 confined flares and 236 CMEs. But for the confined/CME/SEPE problem, we remove 28 events with SEPmax⁡_{\max}=0 in the catalog, indicating uncertain SEPE associations due to overlapping events. The remaining 587 events contain 362 confined flares, 196 CMEs without SEPEs, and 29 SEPEs. Each problem hence uses a mutually exclusive set of event classes. We study three types of models: SXR only, Type-III only, and SXR+Type-III, which use parameters based on SXR and/or type-III properties as their names suggest. Table 4 presents the components of ff for the various models and their performance scores for the two probability-estimation problems. The mean and standard deviation of each score over the 20 repeated five-fold trials are reported.

The SXR+Type-III model provides the best overall scores for both problems. Its BS and CE scores are lower than the SXR-only model, given the estimation errors, demonstrating improved event-occurrence probability estimates. The corresponding event class-predictor gives BA = 0.80 ± 0.005 (0.71 ± 0.015) for the binary (three-class) problem, compared to 0.73 ± 0.003 (0.56 ± 0.009) for the SXR-only model. Thus, DH type-III diagnostics provide information complementary to SXR flare properties in estimating CME and SEPE occurrence probabilities.

A scale-free framework for solar and exospace weather applications

We now define and evaluate the scale-free model, independent of the solar activity scales, and hence should be extendable across stellar flares with varying physical scales while sharing a phenomenological correspondence to solar events. This model also lets us evaluate the space weather relevance of solar and stellar flares within a common flare vector space framework. The model defines ff as a combination of parameters that probe the relative physical scales of non-thermal particle acceleration and heating, explored by radio and SXR observables, respectively. The additional RXR_X is the ratio of FX,pF_{X,p} to the bolometric solar flux at 1 AU, given its effective surface temperature of 5880 K. RXR_X is a common stellar activity metric, known to scale with mean physical parameters, including Rossby number and magnetic field strength [74, 66, 84]. For both probability estimation problems, the scale-free model gives better mean BS, CE, and BA scores than the SXR-only model.

Further, the model calibration plot in Figure 8 compares the space weather class probabilities predicted by the mean-coefficient SXR+Type-III and scale-free models with the observed occurrence fractions for each problem. The calibration plot is constructed as follows. As described in Sect. 5.2.1, each model type and probability-estimation problem yields 100 sets of optimized coefficients, whose mean defines the mean-coefficient model. We apply the mean-coefficient model to the full flare sample and generate a set of predicted probabilities for each SiS_i per flare vector. Then, for each SiS_i, the predicted probability values are divided into bins with equal data counts, and the mean predicted probability and observed event fraction in each bin are plotted. To obtain the prediction uncertainty range, we repeat the above procedure using each of the 100 sets of model coefficients, each time generating the corresponding calibration curve. The shaded uncertainty region shows the 5th-95th percentile range of the 100 resulting calibration curves. This exercise is performed for each model type, probability-estimation problem, and SiS_i. For the binary problem, only the CME class is shown because the confined flare probability and occurrence fraction are complementary to those of the CME class. Both the SXR+Type-III and scale-free models remain close to the perfect calibration line, demonstrating good agreement between the predicted probabilities and observed occurrence fractions. The mean coefficients and their uncertainties for the SXR+Type-III and scale-free models are provided in Appendix § 11. The uncertainties (σ\sigma), estimated as the standard deviation of the set of 100 coefficients, reflect the dependence of the inferred coefficients on the data distribution across the training subsets.

Calibration curves for mean-coefficient SXR+Type-III and scale-free models

Figure 8. Calibration curves for mean-coefficient SXR+Type-III and scale-free models for the two probability-estimation problems: (a) confined/CME and (b) confined/CME/ SEPE. For each problem and event class, the plotted points compare the predicted occurrence probabilities to the observed event occurrence fraction. Shaded regions show the prediction uncertainties. The dashed diagonal denotes perfect calibration. See Sect. 5.2.3 for details. The legend in panel (a) applies to all panels.

These results demonstrate that a Sun-as-a-star scale-free model based on physically motivated parameters derived from SXR and radio observables retains much of the space weather event occurrence probability information, compared to the scale-dependent solar models. This provides a proof of concept for a common scale-free Sun-as-a-star framework for solar and exospace weather assessment. The present scale-free f is minimal and needs future work to extend it meaningfully to higher dimensions by defining physics-relevant multiwavelength Sun-as-a-star flare parameters that probe other independent aspects of plasma and magnetic evolution.

Properties of S-type-III bursts

Only 4% of DH type-II events during 1997–2025 exhibited S-type-III bursts, highlighting their rarity. S-type-III burst events have a median flare class, CME speed, and CME width of M7.9, 1448 km s−1^{-1}, and 360°, respectively. Of the CMEs associated with 26 S-type-III bursts, 81% (21 out of 26) were halo with deprojected space speeds >900 km s−1^{-1}. Space speeds were obtained from the CDAW Halo CME catalog ([33]). Reliable SEP flux measurements, unaffected by co-temporal events, were available for 22 of the 26 burst events. Of these, 17 (77%) bursts were associated with major SEPEs, with fluxes ranging from 10 to 1480 pfu. The 4 out of 5 events without a reported major SEPE are east-side events, of which 1 occurred behind the east limb. The only west-side event without a major SEPE requires a detailed study, beyond the scope of this work, to infer the contributing factors, including the lack of magnetic connectivity of the IP shock regions to the observing spacecraft. Our database provides multi-spacecraft radio, SXR, and SEP flux diagnostics. Interestingly, 13 out of 14 west-side events produced a major SEPE. In general, S-type-III bursts are closely associated with very strong flares with fast halo CMEs that can produce major SEPs. However, their low occurrence rate limits their utility as a general indicator of major SEPEs and CMEs.

Eruptive flares without robust DH type-III features

As mentioned in Sect. 4, most (88%) of the SXR flares without a type-III burst are confined. This can arise from a physical scenario where there are no open field lines connecting the particle acceleration sites to IP space ([25]). We find 16 eruptive flares with extremely weak or missing type-III bursts. Though these events constitute only about 4% (16 out of 375) of the total eruptive flares in our study sample, they are very relevant to stellar CME studies where several radio monitoring studies performed over decades had failed to detect type-II, III, and IV bursts in some very active stars (e.g. [7]; [63]; [80]). The type-III bursts being the most common burst type associated with solar CMEs, the physics of type-III-weak CMEs requires detailed case studies. Some of these events may have metric or kilometric (<1 MHz) type-III bursts without the DH branch due to the details of the magnetic field structure or opacity along the observing line-of-sight, resulting in reduced radio flux. Detailed case studies are beyond the scope of this paper, but our catalog provides multi-spacecraft, multi-waveband diagnostic data products. Here, we will just present an overview of DH type-III-weak eruptive flares with some interesting examples.

The DH type-III-weak CMEs have speeds ranging from 163 to 1200 km s−1^{-1}, and associated flare classes range from C5.4 to M7.4. Three of these CMEs are halo, with two of them associated with DH type-II bursts. Figure 9 shows the dynamic spectrum and 1 MHz light curves from the STEREO and Wind spacecraft, and the GOES SXR flare light curve for these two CMEs. Panel (a) shows the event with the fastest CME and intense DH type-II amongst all eruptive DH type-III weak CMEs. This CME also caused a major SEPE. Strong type-III bursts are detected in the kilometric band, though they cannot be robustly traced in the DH band, especially in the 1 MHz light curve. The bright emission in the 1 MHz Wind data around 03:00 UT is the DH type-II emission. Note that the flare source was behind the limb in the field of view of STEREO A. Figure 9b presents the other halo CME with DH type-II bursts but lacking DH type-III emission. This event also shows kilometric type-III bursts, though it is weaker than the previous event. There was no major SEPE associated with this CME. STEREO A and Wind were separated by just 14∘14^\circ, making the dynamic spectra quite similar, and hence it is not ideal for a robust multi-vantage point event analysis. However, there are several eruptive events in the DH type-III catalog during Cycle 24 when Wind and STEREO spacecraft were well separated to explore the effect of plasma along the line-of-sight and magnetic topology in the observed type-III properties.

Dynamic spectra and light curves for two halo CME events

Figure 9. Examples of halo CME events with DH type-II bursts, but extremely weak or no DH type-III emission. Events with and without an associated major SEPE are shown in panels (a) and (b), respectively.

Conclusion

We performed an extensive search for DH type-III emission associated with all ≥\geqC5-class flares during the period from 2006-12 to 2025-01. Flares with source longitudes within ±55∘\pm55^\circ were only considered. The radio data came from the Wind and STEREO spacecraft. We present a comprehensive catalog of DH type-III bursts and a catalog of flares without robust DH type-III detection. Additionally, we analysed all DH type-II events in the CDAW catalog and compiled a catalog of S-type-III bursts. We provide calibrated DH dynamic spectra, CME height-time plots, links to SXR and white-light movies of the event, and light curves of 1 MHz flux, SXR flux, and proton flux for every event in these catalogs. The CME (speed, width), SXR flare (peak flux [FX,pF_{\mathrm{X,p}}], GOES class, rise time [triset_{\mathrm{rise}}]), and SEPE properties are provided for every event in the catalog.

Of the 734 ≥\geqC5 flare events with good-quality SXR and DH data with no confusion from co-temporal emissions, 538 (73%) show clear co-temporal DH type-III bursts. 196 flares lacked robust DH type-III detections, of which 16 are eruptive. The median DH type-III burst duration (ΔTburst\Delta T_{\mathrm{burst}}), peak flux density (Fr,pF_{\mathrm{r,p}}), integrated flux density (Fr,intF_{\mathrm{r,int}}), and the occurrence time delay of the radio peak flux with respect to SXR (δtpeak\delta t_{\mathrm{peak}}) are 18 min, 1.5×1051.5 \times10^{5} SFU, 3.4×1073.4 \times10^{7} SFU s, and -0.6 min, respectively. We analysed the properties of DH type-III bursts within four event groups in the database, namely confined flares (confined), CMEs without DH type-II (CME-no-DHII), CMEs with DH type-II (CME+DHII), and major SEPEs (SEP). Median values and lower envelopes of every DH type-III property, except δtpeak\delta t_{\mathrm{peak}}, show a systematic increase moving across confined, CME-no-DHII, CME+DHII, and SEP groups. The lower type-III parameter values for confined flares agree with the fact that only a fraction of the accelerated particles may exit the flare site in these events, while most of the released magnetic energy remains confined. This separation between median ΔTburst\Delta T_{\mathrm{burst}}, Fr,pF_{\mathrm{r,p}}, and Fr,intF_{\mathrm{r,int}} values as a function of event group persists even for data within narrow SXR flare-strength bins, indicating that DH type-III burst properties contain information beyond SXR flare strength, which can separate the different event groups. Kolmogorov-Smirnov tests further confirm the statistical distinction of the distributions of these properties, especially ΔTburst\Delta T_{\mathrm{burst}}, for various event group pairs.

Further, we tested whether DH type-III parameters can improve the space weather event assessment beyond SXR parameters, using two occurrence probability-estimation problems: (1) confined/CME and (2) confined/CME/SEPE. For each problem, we derived three event class occurrence probability models differing in their flare parameters: (1) SXR only, (2) Type-III only, and (3) SXR and type-III (SXR+Type-III). We used 20 repetitions of a stratified five-fold cross-validation procedure to derive optimal models in each case, and evaluated the predicted probabilities using Brier score (BS), calibration error (CE) and balanced accuracy (BA). The SXR+Type-III model provides the best overall probability estimates, with BS = 0.13±0.0010.13 \pm0.001 (0.30±0.0030.30 \pm0.003) and CE = 0.03±0.0040.03 \pm0.004 (0.02±0.0030.02 \pm0.003) for the binary (three-class) problem, compared with BS = 0.16±0.0010.16 \pm0.001 (0.37±0.0020.37 \pm0.002) and CE = 0.04±0.0040.04 \pm0.004 (0.04±0.0030.04 \pm0.003) for the SXR-only model. The corresponding class predictors give BA = 0.80±0.0050.80 \pm0.005 (0.71±0.0150.71 \pm0.015), compared with 0.73±0.0030.73 \pm0.003 (0.56±0.0090.56 \pm0.009) for the SXR-only case. The mean-coefficient SXR+Type-III model also closely reproduces the observed space weather event fractions in the full-sample calibration tests. Together, these results demonstrate that DH type-III diagnostics provide information complementary to SXR flare properties for estimating CME and SEPE occurrence probabilities.

The models discussed above are based on Sun-as-a-star parameters that depend on absolute solar flare scales. We further demonstrate that a scale-free Sun-as-a-star model based on DH type-III occurrence, dimensionless ratios of type-III and SXR parameters, and RxR_{\mathrm{x}} outperforms the SXR-only model while remaining well calibrated against the observed space weather event occurrence fractions. This demonstrates that useful event-probability information can be retained without relying solely on absolute solar flare scales. This scale-free model framework can be further improved by incorporating independent plasma-evolution diagnostics from other multiwavelength observables, providing a pathway to extend solar space weather diagnostic tools for exospace weather assessment of stellar flares spanning a wide range of physical scales. Mean model coefficients and their uncertainties for the SXR+Type-III and scale-free models are provided in Appendix § 11.

We find that 26 of 675 DH type-II events (4%) exhibit clear S-type-III bursts. Although rare, 21 of these events were associated with fast halo CMEs with speeds >900 km s−1>900\ \mathrm{km\,s^{-1}}, and 77% were associated with major SEPEs, making S-type-III bursts highly space weather relevant. Meanwhile, 16 out of 375 eruptive flares do not show clear DH type-III emission. We present a few interesting cases of flares with fast halo CMEs but insignificant DH type-III emission, though the type-III features become detectable in the kilometric band. Together with existing multi-spacecraft DH type-II and type-IV catalogs ([28, 57, 60]), the catalogs and statistical framework presented here provide a benchmark for interpreting future DH observations in the context of solar and stellar space weather.

Acknowledgments

We thank the CDAW team for maintaining an up-to-date catalog of flares and space weather events. We thank NASA’s SPDF center for providing science-ready DH band data from Wind and STEREO spacecraft. AM and NG are supported in part by NASA’s STEREO project and LWS program. SA was partially supported by an NSF grant, AGS-2043131. AM acknowledges the developers of the various Python modules listed in the Software section. AM acknowledges Seiji Yashiro for posting the catalog online and maintaining the CDAW data resources.

Author Contributions

AM is responsible for compiling all the catalogs presented here, identifying the DH bursts in the Wind and STEREO databases, doing the analysis, and writing the manuscript. The initial master flare catalog was generated by SA and used by AM to search for radio bursts. SA also cross-matched flare lists with the LASCO-CME catalog. PM is responsible for maintaining the DH type-II, SEP, and GLE event lists. PM also assisted in developing the introduction. NG helped with scientific discussions that improved the manuscript. JB was a short-term intern who briefly assisted with the visual identification of S-type-III bursts.

Facilities: Wind (WAVES), STEREO (SWAVES), SOHO (LASCO), SDO (AIA)

Software: Numpy ([37]), Astropy ([5]), Matplotlib ([43]), Scikit-learn ([65]), Multiprocessing ([52]), Pandas ([64]), Urllib5, Beautifulsoup6, Sunpy ([78]), Datetime7, Scipy ([44])

The various event catalogs

This publication presents 3 catalogs. Two of them concern SXR flares with and without DH type-III detections. Flares with source longitudes within ±55∘\pm55^\circ and occurred between late 2006 and January, 2025 were studied. The third catalog presents all S-type-III bursts associated with DH type-II events in the CDAW catalog covering the period from 1997 to 2025.

DateStartEndDuration (min)log(Peak Flux) (SFU)Integ. Flux (SFUs)DateStartPeaktriset_{\mathrm{rise}} (min)ClassFlux (W/m2^2)SourceTime_wrt_pflr (min)Pk_rankMscoredata_qltyEv_ComplexityVsky (km/s)CME Width (deg)DH type-IISEP Flux (pfu)GLE
2008-12-0718:5219:30335.8110508324.172008-12-0718:2019:1353M3.03.0E-05S06E45-11.6521208301270-1-1
2008-12-1302:1403:15507.7912013109520.602008-12-1302:1402:4026X4.94.9E-04S06W23-8.85212017743600698-1
2009-12-1422:0722:45337.43123758056.472009-12-1421:0722:1568X2.22.2E-04S06W46-0.531201043600215-1
2009-12-2204:5005:07136.08114861719.222009-12-2204:5004:566C7.27.2E-06S26W463.13120318470-1-1
2010-02-0618:5219:15133.41424052.262010-02-0618:4718:5912M4.24.2E-05N21E173.14120240970-1-1
2010-02-0702:2902:58255.2569185821.832010-02-0702:2002:3414M9.92.9E-04N21E107.8541204213600-1-1
2010-02-0803:0803:25115.4238314455.952010-02-0803:0803:179C8.98.9E-06N23E00-1.573120000-1-1
2010-02-0804:0404:27185.68525077494.742010-02-0804:0404:1511M1.11.1E-05N21W010.753120000-1-1
2010-02-0805:1205:30135.45302639193.432010-02-0805:1205:3511M1.21.2E-05N21W01-0.181120153990-1-1
2010-02-0806:0706:1673.918932814.712010-02-0805:5906:067M1.01.0E-05N21W016.554120000-1-1
2010-02-0807:3707:5574.893283614.702010-02-0807:3607:437M6.06.0E-05N26E051.9331202261010-1-1
2010-02-1207:1807:45196.73367524156.342010-02-1207:1807:257M1.11.1E-05N24E13-0.683120000-1-1

Table 10. DH type-III catalog. The details of the radio burst, SXR flare, CME, SEP, and GLE are provided for every event.

DateStartPeakClassFlux (W/m2)Sourcelog(Peak Flux) (SFU)Time_wrt_pflr (min)Pk_rankdata_qltyVsky (km/s)Width (deg)Flux (pfu)
2006-12-0704:2704:45C8.78.7E-06S07E523.181.283254712-1
2007-07-1017:4817:53C5.25.2E-06S07E45nannannan2nannan-1
2010-02-0704:4204:52M1.61.6E-05N20E07nannannan2nannan-1
2010-11-0701:3101:41C7.87.8E-06S22E533.1524.9742nannan-1
2011-02-1404:2904:49M1.21.2E-05S20W01nannannan239044-1
2011-03-1215:1915:29M1.41.4E-05N06W41nannannan217568-1
2011-03-1423:1723:24C1.51.5E-06N18W523.1516.7742nannan-1
2011-04-1600:3400:57C7.57.5E-06N12W273.33-22.6512nannan-1
2011-04-1613:5114:14C8.38.3E-06N13W353.5114.842nannan-1
2011-04-2215:4715:53M1.81.8E-05S18E35nannannan2nannan-1
2011-07-2802:1902:25C7.87.8E-06N18E313.2418.3742nannan-1

Table 11. Catalog of SXR flares with insignificant or no DH type-III. The details of the SXR flare, CME, and SEP are provided for every event. The flux and δtpeak\delta t_{\mathrm{peak}} of the closest DH type-III burst is also provided for relevant events. Note that Wind/WAVES data usually has a background flux level of log F = 3.15 - 3.2 SFU.

Start date [UT]Start time [UT]End date [UT]End time [UT]Start freq [MHz]End freq [MHz]ConfidenceSourceClassPeak Flux (W/m2^2)Date [UT]Time [UT]Speed [km/s]Width [deg]Flux [pfu]
1997-11-0406:0011/0504:30140001001S14W33X2.12.1E-0411/0406:1078536072
1997-11-0612:2011/0708:30140001001S18W63X9.49.4E-0411/0612:101556360495
1998-04-2306:0004/2315:30140002001S17E90X1.21.2E-0404/2305:5516913600
1998-12-1817:5012/1818:151400050000N19E64M8.08.0E-0512/1818:091749360-1
1999-05-0305:5005/0308:4580002001N15E32M4.44.4E-0505/0306:06158436014
1999-06-2218:2506/2218:40300020000N22E37M1.71.7E-0506/2218:541133360-1
2000-02-1720:4202/1822:12140001001S29E07M1.31.3E-0502/1721:3072836010
2000-07-1220:0507/1220:35600010001N19W61M1.51.5E-0507/1220:30820101-1
2000-07-1410:3007/1514:3014000801N22W07X5.75.7E-0407/1410:5416743602400
2000-11-0823:2011/0912:0040002001N10W77M7.47.4E-0511/0823:061738>17014800

Table 12. S-type-III catalog. The details of the DH type-II burst, SXR flare, CME, and SEP are provided for every event.

DH type-III burst properties within flare FX,pF_{\mathrm{X,p}} bins

Mean coefficients for the selected probability estimation models

Table 6 and Table 7 give the mean coefficients and their uncertainty for SXR+Type-III and scale-free models. The values are estimated as described in Sect. 5.2.

ModelParameterβˉ±σβ\bar{\beta} \pm\sigma_{\beta}
SXR + Type-IIIβ0\beta_09.125±1.0539.125 \pm1.053
log⁡FX,p\log F_{\mathrm{X,p}}2.591±0.2062.591 \pm0.206
log⁡trise\log t_{\mathrm{rise}}1.471±0.1891.471 \pm0.189
gg−2.389±0.637-2.389 \pm0.637
glog⁡Fr,pg \log F_{\mathrm{r,p}}0.105±0.1980.105 \pm0.198
glog⁡Fr,intg \log F_{\mathrm{r,int}}0.176±0.2150.176 \pm0.215
glog⁡ΔTburstg \log\Delta T_{\mathrm{burst}}1.918±0.3041.918 \pm0.304
Scale-freeβ0\beta_023.225±1.58523.225 \pm1.585
gg−1.560±0.563-1.560 \pm0.563
glog⁡(Fr,p/FX,p)g \log(F_{\mathrm{r,p}}/F_{\mathrm{X,p}})0.335±0.0550.335 \pm0.055
glog⁡(ΔTburst/trise)g \log(\Delta T_{\mathrm{burst}}/t_{\mathrm{rise}})0.252±0.1660.252 \pm0.166
log⁡RX\log R_{\mathrm{X}}3.141±0.1943.141 \pm0.194

Table 6. Mean model coefficients for the selected Sun-as-a-star SXR+Type-III and scale-free models for the confined/CME probability problem. The reported values are βˉ±σβ\bar{\beta} \pm\sigma_{\beta}. Here, β0\beta_0 is the model intercept, while the remaining coefficients correspond to the respective components of the flare vector.

ModelParameterConfinedCMESEPE
SXR + Type-IIIβ0\beta_0−6.974±0.800-6.974 \pm0.8001.827±0.5391.827 \pm0.5395.146±0.9145.146 \pm0.914
log⁡FX,p\log F_{\mathrm{X,p}}−2.513±0.166-2.513 \pm0.1660.031±0.1100.031 \pm0.1102.482±0.1952.482 \pm0.195
log⁡trise\log t_{\mathrm{rise}}−2.098±0.180-2.098 \pm0.180−0.644±0.161-0.644 \pm0.1612.742±0.2812.742 \pm0.281
gg6.053±0.6476.053 \pm0.6474.126±0.5864.126 \pm0.586−10.179±0.988-10.179 \pm0.988
glog⁡Fr,pg \log F_{\mathrm{r,p}}−0.346±0.149-0.346 \pm0.149−0.292±0.160-0.292 \pm0.1600.638±0.0950.638 \pm0.095
glog⁡Fr,intg \log F_{\mathrm{r,int}}−0.261±0.160-0.261 \pm0.160−0.051±0.160-0.051 \pm0.1600.312±0.0760.312 \pm0.076
glog⁡ΔTburstg \log\Delta T_{\mathrm{burst}}−2.574±0.286-2.574 \pm0.286−0.905±0.328-0.905 \pm0.3283.479±0.5353.479 \pm0.535
Scale-freeβ0\beta_0−21.995±1.239-21.995 \pm1.239−0.337±0.800-0.337 \pm0.80022.332±1.36822.332 \pm1.368
gg3.737±0.4703.737 \pm0.4702.867±0.4022.867 \pm0.402−6.604±0.650-6.604 \pm0.650
glog⁡(Fr,p/FX,p)g \log(F_{\mathrm{r,p}}/F_{\mathrm{X,p}})−0.505±0.047-0.505 \pm0.047−0.226±0.038-0.226 \pm0.0380.731±0.0640.731 \pm0.064
glog⁡(ΔTburst/trise)g \log(\Delta T_{\mathrm{burst}}/t_{\mathrm{rise}})0.497±0.1620.497 \pm0.1620.730±0.1560.730 \pm0.156−1.228±0.263-1.228 \pm0.263
log⁡RX\log R_{\mathrm{X}}−3.064±0.153-3.064 \pm0.153−0.104±0.105-0.104 \pm0.1053.168±0.1783.168 \pm0.178

Table 7. Mean model coefficients for the selected Sun-as-a-star SXR+Type-III and scale-free models for the confined/CME/SEPE probability problem. Entries for each event class give βˉ±σβ\bar{\beta} \pm\sigma_{\beta}; the parameter definitions are the same as in Table 6.

References

  1. [1]Alissandrakis, C. E., Nindos, A., Patsourakos, S., Kontogeorgos, A., & Tsitsipis, P. 2015, A&A, 582, A52, doi: 10.1051/0004-6361/201526265
  2. [2]Aschwanden, M. J., & Freeland, S. L. 2012, ApJ, 754, 112, doi: 10.1088/0004-637X/754/2/112DOI
  3. [3]Aschwanden, M. J., Xu, Y., & Jing, J. 2014, ApJ, 797, 50, doi: 10.1088/0004-637X/797/1/50DOI
  4. [4]Asensio Ramos, A., Cheung, M. C. M., Chifu, I., & Gafeira, R. 2023, Living Reviews in Solar Physics, 20, 4, doi: 10.1007/s41116-023-00038-x
  5. [5]Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068
  6. [6]Aubier, M. G., Leblanc, Y., & Moller-Pedersen, B. 1978, A&A, 70, 685
  7. [7]Bastian, T. S. 1994, SSRv, 68, 261, doi: 10.1007/BF00749152DOI
  8. [8]Bentum, M. J., Verma, M. K., Rajan, R. T., et al. 2020, Advances in Space Research, 65, 856, doi: 10.1016/j.asr.2019.09.007
  9. [9]Benz, A. O., Brajša, R., & Magdalenić, J. 2007, SoPh, 240, 263, doi: 10.1007/s11207-007-0365-9DOI
  10. [10]Bougeret, J. L., Kaiser, M. L., Kellogg, P. J., et al. 1995, SSRv, 71, 231, doi: 10.1007/BF00751331DOI
  11. [11]Browne, M. W. 2000, Journal of Mathematical Psychology, 44, 108, doi: https://doi.org/10.1006/jmps.1999.1279
  12. [12]Cairns, I. H., & Melrose, D. B. 1985, J. Geophys. Res., 90, 6637, doi: 10.1029/JA090iA07p06637DOI
  13. [13]Cane, H. V., Erickson, W. C., & Prestage, N. P. 2002, Journal of Geophysical Research (Space Physics), 107, 1315, doi: 10.1029/2001JA000320
  14. [14]Cane, H. V., Stone, R. G., Fainberg, J., et al. 1981, Geophys. Res. Lett., 8, 1285, doi: 10.1029/GL008i012p01285DOI
  15. [15]Cattell, C., Glesener, L., Leiran, B., et al. 2021, A&A, 650, A6, doi: 10.1051/0004-6361/202039510DOI
  16. [16]Chen, C., Su, Y., Chen, W., et al. 2025, ApJL, 987, L4, doi: 10.3847/2041-8213/addf2fDOI
  17. [17]Cliver, E. W., & Ling, A. G. 2008, 690, 598, doi: 10.1088/0004-637X/690/1/598DOI
  18. [18]DeBoer, D. R., Warnick, K. F., Tremblay, C. D., et al. 2025, in 2025 United States National Committee of URSI National Radio Science Meeting (USNC-URSI NRSM) (IEEE), 46, doi: 10.23919/USNC-URSINRSM66067.2025.10907042DOI
  19. [19]Del Zanna, G., Aulanier, G., Klein, K.-L., & Török, T. 2011, A&A, 526, A137, doi: 10.1051/0004-6361/201015231DOI
  20. [20]Fainberg, J., & Stone, R. G. 1970a, SoPh, 15, 222, doi: 10.1007/BF00149487DOI
  21. [21]Fainberg, J., & Stone, R. G. 1970b, SoPh, 15, 222, doi: 10.1007/BF00149487DOI
  22. [22]Ginzburg, V. L., & Zhelezniakov, V. V. 1958, Soviet Ast., 2, 653
  23. [23]Gopalswamy, N. 2004, in Astrophysics and Space Science Library, Vol. 314, Astrophysics and Space Science Library, ed. D. E. Gary & C. U. Keller, 305, doi: 10.1007/1-4020-2814-8_15
  24. [24]Gopalswamy, N., Aguilar-Rodriguez, E., Yashiro, S., et al. 2005, Journal of Geophysical Research (Space Physics), 110, A12507, doi: 10.1029/2005JA011158
  25. [25]Gopalswamy, N., Akiyama, S., & Yashiro, S. 2009a, in IAU Symposium, Vol. 257, Universal Heliophysical Processes, ed. N. Gopalswamy & D. F. Webb, 283–286, doi: 10.1017/S174392130902941X
  26. [26]Gopalswamy, N., Kumari, A., & Mäkelä, P. A. 2023, in 2023 XXXVth General Assembly and Scientific Symposium of the International Union of Radio Science (URSI GASS), 79, doi: 10.23919/URSIGASS57860.2023.10265368
  27. [27]Gopalswamy, N., Mäkelä, P., Akiyama, S., et al. 2015, ApJ, 806, 8, doi: 10.1088/0004-637X/806/1/8DOI
  28. [28]Gopalswamy, N., Mäkelä, P., & Yashiro, S. 2019, Sun and Geosphere, 14, 111, doi: 10.31401/SunGeo.2019.02.03DOI
  29. [29]Gopalswamy, N., Michalek, G., Yashiro, S., et al. 2026, in IAU Symposium, Vol. 20, IAU Symposium, ed. N. Gopalsway, O. Malandraki, A. Vidotto, & W. Manchester, IV., 133–139, doi: 10.1017/S1743921325000559
  30. [30]Gopalswamy, N., Yashiro, S., & Akiyama, S. 2007, Journal of Geophysical Research (Space Physics), 112, A06112, doi: 10.1029/2006JA012149
  31. [31]Gopalswamy, N., Yashiro, S., Kaiser, M. L., Howard, R. A., & Bougeret, J.-L. 2001, J. Geophys. Res., 106, 29219, doi: 10.1029/2001JA000234DOI
  32. [32]Gopalswamy, N., Yashiro, S., Michalek, G., et al. 2009b, Earth Moon and Planets, 104, 295, doi: 10.1007/s11038-008-9828-7
  33. [33]Gopalswamy, N., Yashiro, S., Michalek, G., et al. 2010, Sun and Geosphere, 5, 7
  34. [34]Gopalswamy, N., Yashiro, S., Xie, H., et al. 2008, ApJ, 674, 560, doi: 10.1086/524765DOI
  35. [35]Gopalswamy, N., Kaiser, M. L., Thompson, B. J., et al. 2000, Geophys. Res. Lett., 27, 1427, doi: https://doi.org/10.1029/1999GL003665
  36. [36]Harra, L., Brooks, D. H., Bale, S. D., et al. 2021, A&A, 650, A7, doi: 10.1051/0004-6361/202039514DOI
  37. [37]Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2DOI
  38. [38]Hawley, S. L., Allred, J. C., Johns-Krull, C. M., et al. 2003, ApJ, 597, 535, doi: 10.1086/378351DOI
  39. [39]Hillaris, A., Bouratzis, C., & Nindos, A. 2016, SoPh, 291, 2049, doi: 10.1007/s11207-016-0946-6DOI
  40. [40]Hirayama, T. 1974, SoPh, 34, 323, doi: 10.1007/BF00153671DOI
  41. [41]Hoerl, A. E., & Kennard, R. W. 1970, Technometrics, 12, 55, doi: 10.1080/00401706.1970.10488634DOI
  42. [42]Howard, R. A., Moses, J. D., Vourlidas, A., et al. 2008, SSRv, 136, 67, doi: 10.1007/s11214-008-9341-4DOI
  43. [43]Hunter, J. D. 2007, Computing in science & engineering, 9, 90
  44. [44]Jones, E., Oliphant, T., Peterson, P., et al. 2001–, SciPy: Open source scientific tools for Python, http://www.scipy.org/
  45. [45]Kahler, S. W. 1982, J. Geophys. Res., 87, 3439, doi: 10.1029/JA087iA05p03439DOI
  46. [46]Kahler, S. W., Cliver, E. W., & Cane, H. V. 1986, Advances in Space Research, 6, 319, doi: 10.1016/0273-1177(86)90168-7
  47. [47]Knapp, M., Paritsky, L., Kononov, E., & Kao, M. M. 2024, https://arxiv.org/abs/2404.08432
  48. [49]Lin, R. P. 1974, SSRv, 16, 189, doi: 10.1007/BF00240886DOI
  49. [52]McKerns, M. M., Strand, L., Sullivan, T., Fang, A., & Aivazis, M. A. 2012, arXiv e-prints, arXiv:1202.1056, doi: 10.48550/arXiv.1202.1056arxiv.org/abs/1202.1056
  50. [53]McLean, D., & Labrum, N. 1985, Solar Radio Astrophysics (Cambridge University Press)
  51. [54]Melrose, D. B. 1970, Australian Journal of Physics, 23, 885, doi: 10.1071/PH700885DOI
  52. [55]Miteva, R., Samwel, S. W., & Krupar, V. 2017, Journal of Space Weather and Space Climate, 7, A37, doi: 10.1051/swsc/2017035DOI
  53. [56]Miteva, R., Samwel, S. W., & Zabunov, S. 2022, Universe, 8, 275, doi: 10.3390/universe8050275DOI
  54. [57]Mohan, A., Gopalswamy, N., Kumari, A., Akiyama, S., & G, S. 2024a, ApJ, 971, 86, doi: 10.3847/1538-4357/ad5315DOI
  55. [59]Mohan, A., Gopalswamy, N., Raju, H., & Akiyama, S. 2024b, A&A, 691, L8, doi: 10.1051/0004-6361/202451072DOI
  56. [60]Mohan, A., Gopalswamy, N., Raju, H., & Akiyama, S. 2026b, 20, 345, doi: 10.1017/S1743921324001558
  57. [61]Mohan, A., Mondal, S., Oberoi, D., & Lonsdale, C. J. 2019, ApJ, 875, 98, doi: 10.3847/1538-4357/ab0ae5DOI
  58. [62]Mohan, A., Mondal, S., Wedemeyer, S., & Gopalswamy, N. 2024c, A&A, 686, A51, doi: 10.1051/0004-6361/202347924DOI
  59. [63]Osten, R. A., & Bastian, T. S. 2006, ApJ, 637, 1016, doi: 10.1086/498410DOI
  60. [64]pandas development team, T. 2020, pandas-dev/pandas: Pandas, latest Zenodo, doi: 10.5281/zenodo.3509134
  61. [65]Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825
  62. [66]Pizzolato, N., Maggio, A., Micela, G., Sciortino, S., & Ventura, P. 2003, A&A, 397, 147, doi: 10.1051/0004-6361:20021560
  63. [67]Reid, H. A. S., & Kontar, E. P. 2017, A&A, 606, A141, doi: 10.1051/0004-6361/201730701DOI
  64. [68]Reid, H. A. S., & Ratcliffe, H. 2014, Research in Astronomy and Astrophysics, 14, 773, doi: 10.1088/1674-4527/14/7/003DOI
  65. [69]Reiner, M. J., & Kaiser, M. L. 1999, Geophys. Res. Lett., 26, 397, doi: 10.1029/1998GL900317
  66. [70]Reiner, M. J., Karlický, M., Jiřička, K., et al. 2000, ApJ, 530, 1049, doi: 10.1086/308394DOI
  67. [71]Rodríguez-García, L., Gómez-Herrero, R., Dresing, N., et al. 2025, A&A, 694, A64, doi: 10.1051/0004-6361/202452158
  68. [72]Saint-Hilaire, P., Vilmer, N., & Kerdraon, A. 2013, ApJ, 762, 60, doi: 10.1088/0004-637X/762/1/60
  69. [73]Shibata, K., Masuda, S., et al. 1995, ApJL, 451, L83, doi: 10.1086/309688
  70. [74]Stepien, K. 1994, A&A, 292, 191
  71. [75]Stewart, R. T., & Labrum, N. R. 1972, SoPh, 27, 192, doi: 10.1007/BF00151783DOI
  72. [76]Stumpo, M., Benella, S., Laurenza, M., et al. 2021, Space Weather, 19, e2021SW002794, doi: https://doi.org/10.1029/2021SW002794
  73. [77]Talebpour Sheshvan, N., & Pohjolainen, S. 2018, SoPh, 293, 148, doi: 10.1007/s11207-018-1371-9
  74. [78]The SunPy Community, Barnes, W. T., Bobra, M. G., et al. 2020, The Astrophysical Journal, 890, 68, doi: 10.3847/1538-4357/ab4f7a
  75. [80]Villadsen, J., & Hallinan, G. 2019, ApJ, 871, 214, doi: 10.3847/1538-4357/aaf88eDOI
  76. [81]Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2DOI
  77. [82]Wild, J. P. 1970, Proceedings of the Astronomical Society of Australia, 1, 365, doi: 10.1017/S1323358000012364DOI
  78. [83]Winter, L. M., & Ledbetter, K. 2015, ApJ, 809, 105, doi: 10.1088/0004-637X/809/1/105DOI
  79. [84]Wright, N. J., Newton, E. R., Williams, P. K. G., Drake, J. J., & Yadav, R. K. 2018, MNRAS, 479, 2351, doi: 10.1093/mnras/sty1670DOI
  80. [85]Yashiro, S., Gopalswamy, N., Michalek, G., et al. 2004, J. Geophys. Res., 109, doi: https://doi.org/10.1029/2003JA010282

Paper details

vlnc:W_6qes5rtk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 4

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents