Use of L5 Data in CME Propagation Models

Introduction

ESA/A. Baker, CC BY-SA 3.0 IGO

Current operational space weather forecasting is largely constrained by availability of assets located at the Lagrange 1 (L1) vantage point, in geocentric orbit or on the ground. These provide effectively a single viewpoint and limit the forecaster’s ability to view solar disk features, such as active regions that, as the Sun rotates, will influence space weather along the Sun-Earth Line (SEL).

When an earthward directed Coronal Mass Ejection (CME) is launched it is observed by near-Earth coronagraphs as a halo signature, which helps to resolve the propagation direction but leaves a degree of ambiguity between the CME width and its radial propagation speed. As a result of these factors, the accuracy of CME forecasting from a single viewpoint is associated with an average uncertainty of typically +/- 12 hours but this belies large spreads such that an individual event may have a forecast with an error more than twice this average figure.

Locating additional instrumentation away from the SEL provides alternative viewpoints that can be combined with SEL measurements to remove or reduce the ambiguities of single viewpoint observations. The Lagrange 5 (L5) vantage point (trailing the Earth by 60 degrees) is ideal from this perspective as it will provide both an early view of solar disc features before they appear on the visible disk as seen from Earth and also a side-on view of Earthward directed CMEs as they propagate out from the Sun.

Study Objectives

The "Use of L5 Data in CME Propagation Models" study is taking a systematic and quantitative approach to assessing the anticipated benefits that the combination of L5 with L1 and near-Earth data will have on operational space weather capabilities. The main focus of the activity is on CME arrival time predications at Earth, as socio-economic studies have demonstrated that these have the largest net impact as a result of their potential to disrupt terrestrial technological infrastructure. However, other space weather factors are also being considered, such as the prediction of high-speed solar wind streams and the profile of Solar Energetic Particle (SEP) events. The main assessment activity will take a step-wise approach, looking at the potential gains from combined L5 and L1 observations in, for example, CME characterization and background solar wind estimation before looking at the overall improvements in end-user products such as forecasts of CME arrival time. Extensive use will be made of data from the NASA STEREO-B spacecraft during the phase of the mission where its orbit placed it close to the L5 location. The work shall take account of the fact that its purpose is for the assessment of improvements in an operational forecasting capability, and will therefore take into account the cadence and timeliness of the input data that is used to drive the models.

Tasks

The activity is split into four sequential tasks each consisting of a set of work packages, many of which run concurrently.

Task-1 Requirements Analysis and Elaboration of Use Cases

The requirements analysis task includes work packages to review the current state of the art, establish whether further elaboration of the SWE Network baseline requirements is needed to support traceability of multi-viewpoint techniques and assessment of the specific Vigil measurement requirements and anticipate product developments based on the baseline payload. A preliminary elaboration of the Use Cases that will be used to quantify and validate the benefits of multi-viewpoint measurements and that will provide the basis for the bulk of the work in the remaining tasks is also developed at this point.

Task-2 Analysis Environment Design and Preparation

This is the preparation task for the Use Case experiments that are scheduled within Task-3. The primary aim of this task is to establish the consistent methodology that will be used for the execution and assessment of the Use Case experiments, to ensure that quantitative comparisons can be made. A key element of this task in the finalisation of the event list that will be used to drive the Use Case experiments taking account of data availability and event characteristics to ensure that they are suitable for supporting the analysis and that considerations such as selection bias are adequately understood. This task also finalises the elaboration of the Use Case experiments which will assess the impact of the inclusion of multi-viewpoint measurements in the analysis pipeline and how measurement properties such as measurement resolution and cadence can affect the quality of the forecast.

Task-3 Multi-viewpoint Assessment

This involves the systematic implementation and application of the specified Use Cases to the agreed set of events and analysis intervals. The list of the Use Cases follows below.

There are necessarily some interactions and inter-dependencies between the Use Cases as shown below. These have been assessed as part of the planning activities and are taken into account to ensure that there is efficient scheduling and exchange of the necessary inputs and outputs.

Task-4 Roadmap for Vigil Service Products and Tools

The final task of the activity will draw together the key findings of the Use Case analysis in order to develop a roadmap for the future utilisation of the Vigil mission data within the context of the SWE services and the exploitation of multi-viewpoint techniques to which Vigil can contribute.

The roadmap will provide recommendations and development steps that will need to be taken in order to build on the potential improvements on CME propagation provided by the L5 data from the Vigil mission. It will also identify capabilities that are not currently mature enough to have been included in the Task-3 assessment activity but have the potential to become operational on the timescale of the Vigil mission.

Multi-viewpoint Assessment Results

To provide a systematic approach to the assessment a common set of CMEs was used across the different Use Cases and experiments (except for UC7 where additional SEP events were needed). Real, rather than simulated events were used to provide the most realistic set of observations. The CMEs selected required multi-viewpoint remote observations in the vicinity of the Sun-Earth-L5 region. These constraints resulted in observations from STEREO-B in 2010/2011 and STEREO-A in 2020/2021. From these periods a selection of 15 CMEs were selected that had clear signatures both in the remote sensing observations and also in near-Earth in-situ from which a “ground-truth” for Earth arrival could be determined.

The list of selected CMEs is shown in the table below, the columns indicating the CME ID, when it was first seen in the coronagraph, the speed for input into the heliospheric propagation models, which instrument observed the CME and whether they were associated with a SEP signature at Earth.

UC1 - CME onset characterisation from multiple vantage points (L1 and L5)

The CME characterisation task made use of coronagraph (SOHO at L1 and STEREO close to L5) and heliospheric imager (STEREO close to L5). A set of experiments were defined corresponding to different data availability & quality scenarios (see table below). Experiment C1 represents the current reference case with only L1 data available. Experiment C3 is the optimal situation with the addition of L5 coronagraph and  heliospheric imager data at their best resolution and cadence. The remaining experiments looked at the impact of removing the HI or reducing the resolution and cadence of the available data.

The analysis was also undertaken using two techniques, the CAT tool cone model (by three separate forecasters) and the Graduated Cylindrical Shell model (GCS, UGOE). These provided the initial near-Sun characterisation of CMEs which is a vital input to the subsequent propagation models. The figure below shows a snapshot of the multi-viewpoint GCS analysis for one of the selected events.

The speed determination for the 15 CMEs is shown in the following plot.  

The results from the assessment indicated the ongoing challenges in using coronagraph and HI data to fit CME initial conditions. For each experiment, the fits produced by the three forecasters sometimes varied widely, even though all three used the CAT tool and were trained to use the CAT tool in the same way. The fact that some CMEs were analysed to be directed away from the Earth (even though in reality they were Earth-directed) also underlines the challenges still present in CME fitting. The GCS results were like the CAT ones, which shows that the issues are general and not related to a specific tool.  There was evidence that CME fits that used L1 data alone had more outliers than those from fits that used the combined L1 + L5 data, and that the L1-only fits produce poorer CME arrival times (as examined more in the CME arrival prediction at L1).

UC2 – CME arrival prediction at L1

The purpose this Use Case was to demonstrate the impact of the range of CME initial conditions produced in the operational prediction of the arrival of CMEs at L1.    The models considered were based either on a full magnetohydrodynamic (MHD) treatment or a simpler empirical drag-based technique and included:   

• WSA Enlil Wang Sheeley Arge and Enlil
• DBEM Drag Based Ensemble Model
• DDM Drag Dynamic Model
• ELEvoHI Ellipse Evolution Model using Heliospheric Imaging

All these models use the CME geometry and kinematics as basic input parameters. The results from the CME characterisation experiments were used to initialise these propagation models to forecast CME arrivals and impact speeds at L1. The forecasts produced by the models were evaluated against the in-situ arrival time and observed sheath arrival speed. Hit, miss, false alarm and correct rejection statistics were determined for the different cases as well as mean, and mean-absolute, errors for the timing and velocity.

The assessment demonstrated the high sensitivity of the results to the initial near-Sun characterisation. Nevertheless, the detailed analysis of the results demonstrated that:

• Hit/miss statistics can be improved when adding L5 data.
  1. The additional L5 view is of high importance for deriving CME directivity and 3D geometry to feed propagation models.
  2. The hit ratio increases and uncertainty decreases when adding information of L5-HI data. However, the longer tracking comes at the cost of lead time. On average HI input is given for CME distances up to 74 solar radii (giving approximately 54h lead-time for the CME prediction).
• Experiments show that the arrival speed errors improve when combining L1+L5+HI data deriving about +/- 12 hours (on average ca. 7 hours less compared to L1 data only). HI data of high-quality in real-time are of utmost importance.

UC3 - Utilisation of magnetograph data from L5 in the generation of boundary conditions for MHD modelling of the background solar wind

The purpose of this part of the study was to produce a series of boundary conditions for running Enlil to demonstrate the impact of L5 magnetograph data in producing improved boundary conditions for MHD models used in operational solar wind prediction. The analysis of the subsequent Enlil output was undertaken in UC6.

Given the absence of pre-existing magnetograph observations near L5, the study was based on simulated observations, exploiting a pre-existing study for this, from already published work. From the reference Sun model simulated L1 and L1+L5 magnetograms, along with three different solar wind models and two input full-sun resolutions were produced. Considering the best for the purpose was the Non-Potential Solar Wind Model. The results on output solar wind shows that, averaging across the 2 solar cycles when L5 observations are included, the error computed for radial velocity decreases by 12% and for radial magnetic field by 11% compared with L1 only.

UC4 - Utilisation of L1 and L5 data in combination to improve / streamline CME ensemble modelling for the purposes of arrival prediction

UC4 was designed to demonstrate that the use of a combination of L1 and L5 data, including mid-course assessment from heliospheric imaging data, improves operational CME ensemble modelling in comparison to the use of L1 data alone. Comparisons were undertaken in three ways:

• Geometric fitting to time elongation profiles provided by the HI instrument on STEREO.
• Using the outputs from the propagation models compared to the HI time elongation profiles to see if the model was leading or lagging compared to the observation.
• Using additional ensemble runs (from HUXt -simplified 1D MHD model , ELEvoHI and DBEM models) and culling ensemble members in order to refine the CME characteristics and L1 arrival forecast.

The utility of HI data in refining CME arrival predictions was examined via four different methodologies, specifically by exploiting HI data from heliocentric distances progressively further from the Sun. In particular, the analyses employ the ecliptic time-elongation profile (“J-maps”) of each of the 15 CMEs, extracted from corresponding ecliptic time-elongation map.

The figure below shows two views of the Heliospheric time-elongation maps used for CME tracking. The same input data is shown for two sensitivity experiments described in UC1 above, the original STEREO-A observations on the right (best case, C3) and the equivalent reduced cadence and resolution on the left (worst case, C5). There was a noticeable degradation in the forecasting capability when using the worst compared to the best case input data.

Brief summaries of the conclusions of the four analysis techniques are as follows:

1. Geometrical Fitting: The three propagation geometrical fitting techniques used were based on the heliospheric imager data alone. All three techniques demonstrated a predisposition for early CME arrival time predictions, when compared to the in-situ determinations. That said, the arrival time accuracy therefrom also improved with increasing extent of the HI time-elongation profiles.
2. HUXt: In most cases, the assimilation of HI data into the Sequential Importance Resampling (SIR) SIR-HUXt CME forecasting methodology improved the arrival time predictions relative to the usage of the basic HUXt model alone. There was an overall tendency for increasing arrival time accuracy with increasing length of time-elongation track assimilated into SIR-HUXt. The assimilation of real HI observations into HUXt, showed significant promise.
3. ELEvoHI: Although based on the HI data, ELEvoHI also required CME initiation characteristics. These parameters could be taken the near Sun characterisations or alternatively derived from HI observations such that the technique relied on HI data alone. For CAT and GCS, arrival time predictions tended to get progressively less accurate for longer and longer tracks, contrary to what would be expected. However, for this population of CMEs, at least, the use of the ELEvoHI forecasting methodology based on CME direction from HI was demonstrated to provide increasingly accurate Earth-arrival time forecasts, albeit accompanied by progressively reduced lead times, out to 50 degrees elongation.
4. DBEM: A rudimentary exercise in pruning DBEM ensemble members based on their proximity to the corresponding HI elongation value showed initial promise in improving CME arrival accuracy, for some CMEs, but its optimisation was beyond the scope of the current project.

UC5 - Data assimilation for enhanced background solar wind and CME prediction utilising L1 and L5 in-situ

Global solar wind models are initialised with photospheric magnetic field observations, but no data assimilation is currently used for solar wind and CME prediction. CMEs are initialised by coronagraph data, but solar wind observations are not used to constrain or improve the forecasts. This part of the study was to quantify the gain in operational solar wind and CME forecast skill from assimilation of L1 and L5 in-situ measurements. The data assimilation made use of the Burger Radial Variational Data Assimilation scheme, which is based upon a simplified, steady-state, solar wind model and enables us to combine coronal model output from HUXt and in situ solar wind speed observations, taking account of the uncertainties in both.
While assessing the improvement in CME arrival time forecasting resulting from in situ solar wind assimilation was challenging, as the effect cannot be easily isolated from uncertainty in the initial CME properties. For the cases used, the average error in CME arrival times was reduced from around 9-10 hours to around 5-7 hours when data from L5 were added to the assimilation.

UC5a - Data assimilation for enhanced background solar wind and CME prediction utilising L1 and L5 in-situ and imagery data

The starting point was the STEREO+CH solar wind forecast model. This is a persistence model using STEREO data off the Sun to Earth line to predict solar wind speed at L1. In addition, it covers the evolution of coronal holes, the sources of solar wind high speed streams, as they rotate over time from STEREO into Earth-view. An expansion/decay of the coronal holes area is found to be related to under-/overestimation of the persistence result at L1. As such, observed changes in the areas of the coronal hole are “compensated” by varying uncertainty ranges in the speed forecast.

The main conclusions from this Use Case are:

• L5 suggests a great potential for solar wind predictions at L1 with relatively high accuracy and large lead times.
• Coronal hole latitude in combination with geometrical considerations based on the location of the Sun, spacecraft and Earth is responsible for differences between SW velocities at L5 and L1. The predictive indicator introduced in this analysis allowed us to account for these differences in advance and further improve the accuracy of predictions. Additionally, data assimilation also improved the predictive skill.
• The results also show the benefit of an EUVI imager onboard Vigil in supplying CH location information.

UC6 - Solar wind feature characterisation and arrival prediction at L1

The purpose of UC6 was to demonstrate the benefits of L5 data on the prediction of solar wind bulk properties with selected metrics and arrival of solar wind features at L1. The output from UC3 was used to run WSA Enlil simulations and then these results were assessed in this Use Case. This UC complements the work undertaken in UC5a.

Time series were compared using three different metrics: Pearson correlation coefficients, RMSE, and Sequence Similarity Factor from the Dynamic Time Warping analysis, which is a technique developed for comparing time series. The results showed that the L5 magnetograms could improve predictions at all solar cycle phases. The differences were relatively small, but the results could be clearer if real magnetograms could be used instead of synthetic ones. The analysis results would also be expected to be more robust if using longer time periods than 5-day predictions.

UC7 - SEP propagation analysis

UC7 was designed to demonstrate the impact of L5 coronagraph and Solar Energetic Particle (SEP) data in improving forecasts of SEP events.

The Solar Particle Radiation (SPARX) software can produce near real-time forecasts of Solar Energetic Particle events following detection of a solar flare of class M or greater. SPARX takes as its input the class of the flare together with its solar coordinates. This Use Case was run in a subset of the selected CME sample listed. This part of the study was used to give an indication ofthe usage of the SPARX forecasting for different sensitivity experiments, by taking into account CME velocities and widths measured from L1 and from both L1 and L5.

When all the studied events were considered, the results could be summarised as follows:

1. It was found that some events were poorly magnetically-connected and would not normally be expected to produce SEPs visible at Earth and may not have been the source of the observed enhancement.
2. For the other events, the improvement in the SPARX forecast when CME velocity was taken into account was clear and improved further by the inclusion of CME width.
3. For the SEP peak time forecast, the original configuration of SPARX performed rather poorly but the best results were obtained by the inclusion of CME velocity and width data.
4. For SEP event duration, SPARX struggled. In its original configuration only two forecasts were possible, and both of these were very inaccurate.

In its original configuration, SPARX performed quite poorly in forecasting these events. However, the errors were significantly reduced when CME data were included. The results for the forecast of the SEP peak time were also improved, albeit not by as much as for the onset time. Event duration was still a challenge for SPARX.

Overall Use Case Assessment Summary

They key points can be summarised as follows:

CME detection and forecasting:

• Use of coronagraph and HI data for CME detection and forecasting:
  • Estimation of CME initial conditions from these data remains challenging– the CAT fits produced by the three MOSWOC forecasters sometimes varied widely, and some CMEs were analysed to be directed away from the Earth (even though in reality they were Earth-directed). A subset of the experiments characterised using the GCS technique did not show a systematic improvement over the simpler CAT analysis.
  • Having said that, the uncertainty was still sufficiently small that we could reproduce a well-known result – that using the additional L5 coronagraph view in addition to the L1 view improved CME fits and forecasts.
  • The impact of adding HI data to the CME fits was overall limited. Furthermore, the CAT is not really optimised for the combined use of coronagraph and HI and further tool improvement is needed. There were also some issues with for slow events where the selected data time window used for forecasters was too short to see evolution of the event into the HI field of view.
  • In general, there was little clear sensitivity to using “best” or “worst” image cadences or resolution, which is likely related to the uncertainty in fitting mentioned above.
  • However, for CME forecasts which used HI data for both CME initial conditions and CME tracking (ELEvoHI) the benefit of using higher cadence and higher resolution data was shown. The conclusion from these experiments is that the quality of HI image data on which the tracking is performed is of higher importance than the GCS/CAT reconstruction quality.
• Use of L5 in situ data for data assimilation
  • The impact of this assimilation reduced ambient wind errors and led to a reduction in the average error in CME arrival times from around 9-10 hours to around 5-7 hours. For real progress for operational CME forecasting, errors in the initial CME properties also need to be reduced
  • It was also pointed out that using EUV image data from L5 might improve the representation of CME initial conditions.

Impacts on ambient solar wind prediction:

• Use of L5 magnetographs in addition to L1 magnetographs results showed that the L5 magnetograms could improve predictions at all solar cycle phases
• Use of assimilated in situ data from L5 reduced the error by 15% compared to experiments which assimilated L1 data alone
• UC5a also showed the benefits of using in situ data to improve the representation of the ambient wind. The results also show the benefit of an EUVI imager onboard Vigil in supplying CH location information.

Impact on SEP forecasts:

1. Use of CME width information improved SEP onset time metrics and use of data from both L1 and L5 viewpoints reduced the MAE and standard deviation to negligible values. This indicates the benefit that L5 information (in this use case, from coronagraph images) can have on SEP forecasts.

The overall assessment is that the ability to improve operational forecasting of CME arrival is currently constrained by the ability to accurately characterise the near-Sun parameters that are then used to drive the heliospheric propagation models. The additional L5 viewpoint helps in this regard compared to the single head-on view from L1. However, further improvements on the current operational baseline methodologies and techniques are needed to fully realise these benefits. Improvements in the background solar wind estimation through which the CMEs propagate via data assimilation and machine learning have been demonstrated, as have the use of HI data to provide mid-transit tracking of the CME from (as shown by the ELEvoHI results). Currently these improvements do not feed back into reassessment and reanalysis of the near-Sun characteristics. Such a holistic approach is needed to address the constraints of the current serial approach to operational forecasting pipeline.

Study Team

The study is being undertaken by an experienced team of space weather model developers, researchers, service providers and forecasters.