Session 6 - Unveiling Current Challenges in Space Weather Forecasting
Anastasios Anastasiadis (National Observatory of Athens, IAASARS), Enrico Camporeale (Centre for Mathematics and Computer Science, CWI), Manolis K. Georgoulis (Academy of Athens, RCAAM), Ryan McGranaghan (Jet Propulsion Laboratory)
Wednesday 7/11, 09:00-10:30 & 11:15-12:45
MTC 00.10, Large lecture room
Predicting the conditions of our space environment is a true challenge, due to the large size of the system and the complex interplay of physical mechanisms. Nowadays, forecasting techniques range from physics-based to data-driven statistical models. Massively expanded data availability and sophisticated means to analyze voluminous and complex information open new possibilities to innovative methodologies. This session is devoted to the broad spectrum of advanced forecasting techniques, including physical models, statistical methods, data assimilation, information theory, and machine learning. The goal of this session is to provide a forum for new and ongoing efforts that connect the dots between space weather research and future operational forecasting applications. We invite abstracts covering observations, models, and their combinations. Methods that use innovative and multidisciplinary approaches are particularly welcome.
Click here to toggle abstract display in the schedule
Talks : Time scheduleWednesday November 7, 09:00 - 10:30, MTC 00.10, Large lecture room| 09:00 | The Solar Drivers of Space Weather: Review of the Open Issues Affecting Forecasting | Vourlidas, A et al. | Invited Oral | | | Angelos Vourlidas | | | JHUAPL | | | There has been considerable research progress on the solar drivers of Space Weather (SpW), i.e. Coronal Mass Ejections (CMEs), flares, and solar energetic particles (SEPs), thanks to a plethora of space missions and modeling advances in the last 20 years. For example, the inner heliospheric evolution of CMEs is no longer a mystery, thanks to the STEREO observations since 2007. The initiation of CMEs and flares can now be observed with cadence of tens of seconds with arc second resolution, thanks to SDO. The flow of energy required to power solar eruptions is beginning to be understood, thanks to Hinode and IRIS. Muti-viewpoint observations are beginning to unravel rhe evolution of CME shocks and longitudinal distributions of SEPs.
Yet, there has been relatively little progress in forecasting the geoeffectiveness of a particular CME and associated flares and SEPs. Why is that? What are the issues that holding back progress in medium-term forecasting of Space Weather?
I discuss some of the open issues and possible mitigation strategies in this talk.
| | 09:25 | On the usage of Principal Components Analysis (PCA) for the Prediction of Solar Energetic Particle (SEP) events | Papaioannou, A et al. | Oral | | | A. Papaioannou[1], A. Anastasiadis[1], A. Kouloumvakos[2], M. Paassilta[3], R. Vainio[3], E. Valtonen[3], A. Belov[4], E. Eroshenko[4], M. Abunina[4], A. Abunin[4] | | | [1] Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing (IAASARS), National Observatory of Athens, Greece, [2] IRAP, Université de Toulouse, CNRS, CNES, UPS, Toulouse, France, [3] Department of Physics and Astronomy, University of Turku, 20014 Turku, Finland, [4] Institute of Terrestrial Magnetism, Ionosphere and Radiowave Propagation by N.V. Pushkov RAS (IZMIRAN), Moscow Troitsk, Russia | | | Solar Energetic Particle (SEP) events and their parent solar events (e.g. solar flares - SFs and coronal mass ejections - CMEs) are closely related. A wealth of statistical studies has indicated the dependence of the probability of occurrence of SEP events on the magnitude and the longitude of the SF, as well as the velocity and the width of the CME. However, most studies are limited to two dimensional correlations. The situation is further complicated by the fact that the solar parameters are not independent. In this work, we perform a Principal Component Analysis (PCA) on a set of six (6) solar variables (i.e. CME width/size (s) and velocity (u), logarithm of the solar are (SF) magnitude (logSXRs), SF longitude (Lon), duration (DT) and rise time (RT)). In our analysis, we utilize 126 SEP events with complete solar information, from 1997 to 2013. Each SEP event is a vector in six dimensions (corresponding to these six solar variables). We show a classification of the SEP event radiation impact (in terms of the NOAA scales) with respect to the characteristics of their parent solar events. We further attempt to infer the possible prediction of SEP events, applying logistic regression analysis, downstream of the PCA, with single, as well as, multiple explanatory variables. Our goal is to develop a new index (I) for the nowcasting (short-term forecasting) of SEP events. We test several different schemes for I and validate our findings with the implementation of categorical scores (Probability of Detection - POD, False Alarm Rate - FAR). We present and interpret the obtained scores and we discuss the strengths and weaknesses of the different implementations. It is shown that I holds prognosis potential for SEP events, with the maximum POD achieved being 77.78% and the relative FAR, 40.96%.
AP would like to acknowledge support from a post-doctoral IKY scholarship funded by the action "Supporting post-doctoral researchers" from the resources of the b.p. "Human Resources Development Education and Lifelong Learning" with Priority Axes 6,8,9 and co-funded by the European Social Fund and the Greek government. AA would further like to acknowledge the "SPECS: Solar Particle Events and foreCasting Studies" research grant of the National Observatory of Athens. MP and RV acknowledge the funding from the Academy of Finland (decision 267186). Research conducted by MP and RV was further supported by ESA contract 4000120480/17/NL/LF/hh.
| | 09:40 | Challenges of space weather and space radiation predictions for human space explorations | Guo, J et al. | Oral | | | Jingnan Guo[1], R F Wimmer-Schweingruber[1], Donald M. Hassler[2], Cary Zeitlin[3], Bent Ehresmann[2] | | | [1]IEAP, University of Kiel, Kiel, Germany, (2) Southwest Research Institute, Boulder, CO, USA (3) Leidos, Houston, Texas, USA | | | Space agencies such as ESA, NASA, the Chinese space agency and even private companies are launching new human deep space exploration programs to the Moon and Mars. Radiation is one of the most important long-term risks to such missions. In preparation, this requires a very timely and thorough study to better understand the space weather conditions and their effects as a baseline for the development of mitigation strategies against radiation risks.
Radiation damage in space comes mainly form two sources, Galactic Cosmic Rays (GCRs) and Solar Energetic Particles (SEPs). The GCR is omnipresent, ubiquitous, and increases the chance of long term health consequences, such as onset of cancer, cardiovascular diseases, skin atrophy, eye cataract, leukemia, anemia, leucopenia and malfunctions of the central nervous system.
On the other hand, intense SEP events can result in very high dose rates that may exceed the threshold for acute radiation syndrome (ARS) or sickness or poisoning or toxicity after a whole body exposure to high doses of radiation at the Gy [J/kg] level. Such events, despite of being rather infrequent, are very unpredictable and could result in severe damage to humans and equipment and lead to failure of the entire mission. It is therefore especially important to plan mitigation against their effects, for instance by providing a “storm shelter” for astronauts, and using multiple redundant systems.
Under different shielding environment, the intensity and composition of the GCRs/SEPs may vary due to the interactions of primary particles (of different energies and charges) with the surrounding material (such as the spacecraft and the planetary atmosphere) and the generation of secondaries. Habitable shelters on the Moon and Mars with regolith shielding could provide sufficient protection against such radiation hazards. However the situation of a transit spacecraft in deep space or an astronaut carrying out extra-vehicle or planetary surface activities may be much more severe, especially during the onset of a solar particle event (SPE). This is because SPEs generally have a sudden and sporadic nature and can be very intense, dynamic and vary drastically in time and location. Therefore radiation and particle enhancements measured near Earth (such as the onset time and intensity) may be completely different from of that detected elsewhere in the heliosphere such as on the surface Mars.
Radiation exposure from Solar Particle Events is the result of three major procedures: (1) the acceleration process at the Sun which are often related to the flare eruptions and associated shocks, (2) the properties of the accelerated particles injected into the open magnetic fields which are connected to the missions (that can be very differently connected compared to Earth), and (3) the atomic and nuclear interactions of particles with the local shielding environment (such as the spacecraft or the Martian atmosphere). Taking into consideration of these 3 factors, we will show our recent study of the September 2017 event which is seen on the surface of both Earth and Mars as well as at STEREO-A (a spacecraft surrounding the Sun at 1 AU). These three locations have a heliospheric longitudinal separation of more than 240 degrees apart and they all saw the SPE with different time profiles, particle spectra and radiation intensities. We highlight the utmost importance of utilizing multi-spacecraft in-situ and remote sensing observations of the Sun and the heliosphere to better understand such dynamic events and their dynamic effects across the heliosphere in particular at locations where human explorations may take place. | | 09:55 | Efficient photospheric flow tracking and machine learning for physics-based characterization of the solar activity | Attie, R et al. | Oral | | | Raphael Attie, Michael Kirk, Barbara Thompson | | | NASA GSFC | | | Since the year 2010, SDO has sent nearly 3 petabytes of solar observations. By providing such an unprecedented large and varied data sets, this mission has propelled the solar physics community into the era of “Big Data” analytics. To address this new technical and scientific challenge, we present here an innovative three-part framework for efficient data mining and analysis of the solar activity using SDO/HMI:
(i) A method for tracking the horizontal photospheric flows uses an improved version of “Balltracking”: We present the most recent version of this flow tracking algorithm, its advantage over other more traditional methods like Local Correlation Tracking (LCT), and how it has been specifically tuned to handle the massive datastream of SDO/HMI.
(ii) A method for tracking the magnetic flux using HMI data called “Magnetic Balltracking”: We show how it enables us to accurately track magnetic elements on magnetograms in the Lagrange reference frame and systematically derive parameters such as the position, velocity, and fragments area, while providing a univocal identification of all tracked fragments and how we use it to automate the detection of flux emergence.
(iii) An algorithm for segmenting the magnetic flux and flow field: Our framework offers an unprecedented view of the evolution of active regions. We will present a new way to perform feature engineering for Machine Learning as our new physics-based descriptors are used as input of Support Vector Machine for classification and prediction purposes.
Through this multidisciplinary approach, we will show how this expands our knowledge-base on the evolution of active regions, and how it augments our ability to forecast the solar activity. | | 10:10 | Forecasting Solar Wind Velocities from Coronal Hole Properties using Machine Learning Techniques. | Garton, T et al. | Oral | | | Tadhg Garton[1], Joan Campanya[1], Peter Gallagher[1], Sophie Murray[1], David Jackson[2] | | | [1] Trinity College Dublin, [2] UK Met Office, Exeter | | | Solar coronal holes (CH) are regions of open magnetic fields that appear as dark areas on the solar disk in extreme ultra-violet pass-bands. These regions are associated with the high speed solar wind. We predict the solar wind speed at L1 with correlation coefficient of ~0.7 between measured and forecast solar wind, and define the main variables and areas of interest on the solar disk for solar wind forecasting. Solar wind data was measured by the Advance Composition Explorer, CH data is extracted by the CHIMERA algorithm from Solar Dynamics Observatory images, and the forecast was performed using machine learning techniques, such as random forest regression, support vector regression, and neural networks. Inputs to the machine learning algorithm include CH area, CH magnetic field properties and CH disk locations. This improvement to solar wind speed forecasting shows the value of machine learning for forecasting space weather. |
Wednesday November 7, 11:15 - 12:45, MTC 00.10, Large lecture room| 11:15 | Applied artificial intelligence for space weather research | Guhathakurta, M et al. | Invited Oral | | | Madhulika Guhathakurta | | | Exploration Technology Directorate, NASA ARC | | | The recent advances in Artificial Intelligence (AI) capabilities are particularly relevant to NASA Heliophysics because there is growing evidence that AI techniques can improve our ability to model, understand and predict solar activity using the petabytes of space weather data already within NASA archives. This represents a strategic opportunity, since the need to improve our understanding of space weather is not only mandated by directives such as the National Space Weather Action Plan and the Presidential Executive Order for Coordinating Efforts to Prepare the Nation for Space Weather Events, but also because space weather is a critical consideration for astronaut safety as NASA moves forward with the Space Policy Directive to leave LEO and return to the Moon.
The Frontier Development Lab (FDL) is an AI research accelerator that was established in 2016 to apply emerging AI technologies to space science challenges which are central to NASA's mission priorities.
FDL is a partnership between NASA Ames Research Center and the SETI Institute, with corporate sponsors that include IBM, Intel, NVidia, Google, Lockheed, Autodesk, Xprize, Space Resources Luxembourg, as well as USC and other organizations. The goal of FDL is to apply leading edge Artificial Intelligence and Machine Learning (AI/ML) tools to space challenges that impact space exploration and development, and even humanity.
The applied AI projects for space weather that are being undertaken by the Frontier Development Lab (FDL) represent an ideal opportunity for utilization of vast amount of NASA and other data to leverage the public-private partnerships of the FDL program in a manner that is highly complementary to ongoing efforts in space weather research. In this talk I will summarize the findings from two space weather topics, “Solar flare forecasting” and “A tool for exploring variability of Solar-Terrestrial interactions” that were part of FDL 2017 summer program and also the topics from FDL 2018 under way.
| | 11:40 | Recurrence analysis of the magnetospheric dynamics | Consolini, G et al. | Oral | | | G. Consolini[1], T. Alberti[1], P. De Michelis[2], M. Laurenza[1], M.F. Marcucci[1], R.V. Donner[3] | | | [1]INAF – Istituto di Astrofisica e Planetologia Spaziali, via del Fosso del Cavaliere 100, 00133 Roma, Italy, [2]Istituto Nazionale di Geofisica e Vulcanologia, via di Vigna Murata 605, 00143 Roma, Italy , [3]Potsdam Institute for Climate Impact Research, Telegrafenberg A31, 14473 Potsdam, Germany | | | The Earth's magnetosphere displays a nonlinear, multiscale and complex dynamics in response to changes of the solar wind and interplanetary medium conditions. This complex dynamics, which manifests in scale-invariant features of geomagnetic indices and energy relaxation events, has been associated with the dynamical changes occurring commonly in the near and mid magnetotail regions. These features, which appear during geomagnetic storms and substorms, resemble those observed in non-equilibrium phase transitions. Analysing the Earth’s magnetospheric response during some recent geomagnetic storms, Alberti et al. [1] showed that a clear separation of timescales exists between the internal and externally-driven magnetospheric dynamics.
The purpose of this study is to quantify the number and duration of recurrences of magnetospheric variability patterns by investigating both Sym-H and AE indices phase space trajectories on different characteristic timescales by applying the Recurrence Quantification Analysis (RQA). Results confirm the existence of a timescale separation between short (<200 min) and long-term (>200 min) dynamical components with different features. While short-term dynamics is characterized by complicated recurrence patterns exhibiting various types of structures, indicating complex variability (i.e., manifesting a chaotic behavior), the long-term one seems to be characterized by a regular dynamics as a response to external forcing.
The relevance of our results is discussed in the framework of Space Weather modelling and forecasting.
References
[1] Alberti, T., et al., Timescale separation in the solar wind-magnetosphere coupling during St. Patrick's Day storms in 2013 and 2015, J. Geophys. Res. Space Physics, 122, 4266, 2017, doi:10.1002/2016JA023175
| | 11:55 | Long-term electron radiation belt data assimilation based on multiple spacecraft | Cervantes villa, J et al. | Oral | | | I. Pelivan[1], J.S. Cervantes Villa[1], Y.Y. Shprits[1][2][3], A.C. Kellerman[3], A.Y. Drozdov[3], J. de Wiljes[2], S. Reich[2] | | | [1] Helmholtz Centre Potsdam, GFZ, German Research Centre For Geosciences, 14473 Potsdam, Germany, [2] University of Potsdam, 14469 Potsdam, Germany, [3] Department of Earth, Planetary, and Space Sciences, University of California, Los Angeles, California 90095, USA | | | Data assimilation, the combination of observations with models, has become increasingly important in the field of Earth’s radiation belt research. Bringing sparse observations into context with a global model helps to better understand the underlying physics observed during quiet and storm times. This context also allows spacecraft operators to evaluate if an anomaly has been caused by the environment. With real-time data available to correct the model state in particular during enhanced geomagnetic activity, an improved global radiation belt forecast is achieved, indicating whether damage precaution steps may be necessary to protect spacecraft instruments or human infrastructure such as high-latitude power grids.
We present a framework that allows assimilation of different data sources from multiple spacecraft into the 3D-VERB code for a time span of more than 30 years covering the radiation environment down to LEO based on available data. The framework by default uses the validated split-operator sequential Kalman filter [1]. The framework however allows to implement and test other methods such as the popular ensemble Kalman filter. Blending data from very different sources with VERB, we address challenges such as the large span in order of magnitude when combining measurements of LEO and MEO spacecraft which leads to ignoring observed fluxes of comparatively low magnitude in standard assimilation methods.
A long-term radiation belt reanalysis allows to quantify which filter methods perform best in being reliable, accurate and usable for operational forecast. Furthermore, a long-term dataset of globally reconstructed fluxes will be beneficial to improve the understanding of global and regional radiation belt dynamics. For example, the dropout of radiation belt electron fluxes – a drastic variation in the Earth magnetosphere – can be studied over several solar cycles. The reanalysis data can also be used to improve models w.r.t. various phenomena, e.g. in finding a parameterization for the sporadic electromagnetic ion cyclotron (EMIC) waves which play an important role in the dynamics of ultrarelativistic electron population [2].
[1] Shprits, Y., A. Kellerman, D. Kondrashov, and D. Subbotin (2013), Application of a new data operator‐splitting data assimilation technique to the 3‐D VERB diffusion code and CRRES measurements, Geophys. Res. Lett., 40, 4998–5002, doi: 10.1002/grl.50969.
[2] Drozdov, A. Y., Y. Y. Shprits, M. E. Usanova, N. A. Aseev, A. C. Kellerman, and H. Zhu (2017), EMIC wave parameterization in the long‐term VERB code simulation, J. Geophys. Res. Space Physics, 122, 8488–8501, doi: 10.1002/2017JA024389. | | 12:10 | Multiple hours ahead forecast of the Dst index using a combination of Long Short-Term Memory neural network and Gaussian Process | Gruet, M et al. | Oral | | | Marina Gruet[1], Mandar Chandorkar[2], Angélica Sicard[1], Enrico Camporeale[2] | | | [1]ONERA, [2], Center for Mathematics and Computer Science (CWI), | | | In this study, we present a method that combines a Long Short-Term Memory (LSTM) recurrent neural network with a Gaussian Process (GP) model to provide up to 6-hour ahead probabilistic forecasts of the Dst geomagnetic index. The proposed approach brings together the sequence modelling capabilities of a recurrent neural network with the error bars and confidence bounds provided by a Gaussian process. Our model is trained using the hourly OMNI and GPS databases, both of which are publicly available.We first develop a Long Short-Term Memory network to get a single point prediction of Dst. This model yields great accuracy in forecasting the Dst index from 1h ahead to 6h ahead, with a correlation coefficient always higher than 0.873 and a root mean square error lower than 9.86. However, even if global metrics show excellent performance, it remains poor in predicting intense storms (Dst<-250 nT) 6 hours in advance. To improve it and to obtain probabilistic forecasts, we combine the LSTM model obtained with a Gaussian process, and evaluate the hybrid predictor using the Receiver Operating Characteristic curve and the reliability diagram. We conclude that this hybrid methodology provides improvements in the forecast of geomagnetic storms, from 1 hour to 6 hours ahead. | | 12:25 | How predictable are the polar ionospheric equivalent currents? | Freeman, M et al. | Oral | | | R. Shore[1], M. P. Freeman[1], J. W. Gjerloev[2,3], J. C. Coxon[4], E. G. Thomas[5], N. Olsen[6] | | | [1]Space Weather and Atmosphere, British Antarctic Survey, Cambridge, UK, [2]Department of Physics and Technology, University of Bergen, Norway, [3]Johns Hopkins University-Applied Physics Laboratory, Laurel, MD, USA, [4]Space Environment Physics, University of Southampton, UK, [5]Thayer School of Engineering, Dartmouth College, Hanover, USA, [6]DTU Space, Technical University of Denmark, Kongens Lyngby, Denmark | | | Shore et al. (2018)* have used empirical orthogonal function (EOF) analysis to decompose the northern polar surface magnetic field variability in space and time into a small set of characteristic patterns – basis vectors defined in space and time – which collectively describe the majority of the variance. The resulting magnetic field model describes the surface magnetic field variation above 50 degrees North between 1997 and 2009 and represents the complex interactions of local time, season, ionospheric conductivity, and solar and geomagnetic activity. Using this model, we regress the surface magnetic field variation against time series of solar wind parameters in order to obtain covariance maps in magnetic latitude and magnetic local time of the contributions to the ionospheric equivalent current variability from specific solar wind inputs. The following parameters are considered: epsilon, interplanetary magnetic field (IMF) Bz and By components, dipole tilt angle and F10.7 solar flux. We also investigated the contributions from solar wind speed and proton density, though these were not important stand-alone predictors.
Specifically, whilst Shore et al. (2018) computed the orthogonal set of equivalent current patterns of maximal variance which describe the data, here we extract from the combined EOF patterns the spatial patterns due to forcing by each solar wind direct-driving component. We compute these covariance maps as a function of the time lag between the solar wind perturbation and the terrestrial response, demonstrating (with the best fidelity to date) the reconfiguration timescale of the ionospheric equivalent currents to perturbations in different solar wind parameters. Our study is the first to resolve the contributions to the ionospheric equivalent current variability from given IMF components. This new information will improve estimates of the geoeffectiveness of a given solar wind perturbation.
* Shore, R. M., Freeman, M. P., & Gjerloev, J. W. (2018). An empirical orthogonal function reanalysis of the northern polar external and induced magnetic field during solar cycle 23. Journal of Geophysical Research: Space Physics, 123, 781–795. https://doi.org/10.1002/2017JA024420
|
Posters| 1 | Prediction Model for Ionospheric Total Electron Content Based on Deep Learning Recurrent Neural Network | Tianjiao, Y et al. | p-Poster | | | Tianjiao Yuan, Yanhong Chen, Siqing Liu, Jiancun Gong | | | | | | A 24 h ahead forecasting model for ionospheric total electron content (TEC) at Beijing station is established based on the deep learning recurrent neural network (RNN) for the first time. The model implementation requires solar 10.7 cm flux index, geomagnetic index ap, grid map of TEC, solar wind speed and the southward components of interplanetary magnetic field. The predicting results for Beijing station (40°N, 115°E) show that the root mean square error (RMSE) of the disturbed ionosphere TEC predicted by RNN model is lower than that of BPNN (Back Propagation Neural Network) model by 0.49~1.46 TECU. The forecasting accuracy of ionospheric positive storm by RNN model is increased by 16.8% with solar wind parameters. Furthermore, the RMSE of RNN model on 31 strong TEC storm in 2001 and 2015 are less than that of BPNN model by 0.2 TECU, and the RMSE of RNN model is decreased by 0.36~0.47 TECU as solar wind parameters are added. The results indicate that RNN model is more reliable than BP model for short-term forecasting of TEC. Moreover, the add of interplanetary solar wind parameters are helpful for predicting TEC positive storm. | | 2 | Poynting flux evolution in active regions | Chicrala, A et al. | p-Poster | | | Andre Chicrala, D. Shaun Bloomfield | | | Northumbria University, UK | | | Evolution of solar active regions (ARs) is directly related to the occurrence of flares and coronal mass ejections. In this sense, changes in AR magnetic field can be used to unveil other relevant features like plasma flows in the region. Here, observations of surface 3D vector magnetic field from SDO/HMI are studied with the differential affine velocity estimator for vector magnetograms (DAVE4VM) to recover 3D plasma flows in the photosphere. The evolution of vertical Poynting flux (separated into components due to magnetic flux emergence and horizontal motions) is presented for different active regions that produce differing levels of flaring activity over the time they were observed. | | 3 | Forecasting of a Solar Wind Classification using Convolutional Neural Networks | Depypere, G et al. | p-Poster | | | Gilles Depypere[1], Jorge Amaya[2], Giovanni Lapenta[3], Savvas Raptis[4], Adam Shamash[5] | | | [1],[2],[3],[4],[5] KU Leuven | | | \textbf{Introduction:} The Earth is indirectly influenced by what happens on the Sun. Coronal Mass Ejections, Flares or even the steady Solar Wind coming from the Sun is able to influence the Earth by injecting particles into its atmosphere. This might irradiate aircraft passengers, cause interference in communication media or induce electric charges in spacecrafts, satellites and long distance electrical wires. This may have vast economic consequences. The field of Space Weather tries to forecast these events as good as possible but current state-of-the-art models such as WSA-ENLIL+cone still have considerable errors. In this work, a new approach will be tested to create a model to forecast a Solar Wind classification using Convolutional Neural Networks (‘CNN’).
\textbf{Method:} For each hour, the Camporeale classification divides the Solar Wind in four different categories based on seven variables. The classification has the format of four numbers between zero and one, denoting the probability that the solar wind is in one of the four classes. A 2D-image input has been created, linked to this output through a long term mean of the solar wind, to forecast this classification in the best possible way. The input is created with information from SOHO EIT, the GOES x-ray detector and the CACTus CME database. Different CNN architectures with different hyperparameters will be tested linking this input and output to each other in order to create a model that is able to forecast the output from the input. Making the input compatible with SDO data will make real-time forecasts possible.
\textbf{Result:} The coefficient of determination (‘R²’) compares the quality of the model to the mean of the data, with a value of one denoting a perfect fit of the model. The results are very dependent on the time period that is chosen to forecast, with some years producing an R²-value of about 0.44 while other years produce an R²-value of 0.02. Different architectures and hyperparameters produce comparable results.
\textbf{Discussion:} The obtained values for the coefficient of determination show that the model has insight in the input and is able to forecast a significant part of the output. Since the CNN is already able to catch a notable share of the cause-and-effect structure between input and output, using other input and/or output could improve the quality of the model even more.
This work is part of the AIDA H2020 (www.aida-space.eu) project funded by the European Community (EC). | | 4 | Deep learning approach to next-day forecast of solar wind parameters at L1. | Shneider, C et al. | p-Poster | | | Carl Shneider[1], Mandar Chandorkar[1], Enrico Camporeale[1] | | | [1] Centrum Wiskunde & Informatica (CWI) | | | Deep learning has proven extremely successful both in classification and regression problems, especially when it is trained on very large datasets. In the Space Weather context, despite the unarguably large amount of data at our disposal, it remains an open question whether historical datasets contain enough information to build a predictive deep learning system. In this work, we use multiwavelength solar images from both SOHO (Solar and Heliospheric Observatory) and SDO (Solar Dynamics Observatory) as inputs to a convolutional neural network, to predict solar wind parameters observed at L1, 24 hours ahead. This work is part of the AIDA H2020 (www.aida-space.eu) project funded by the European Commission (EC). | | 5 | Statistical validation of an empirical model of solar proton event time profiles | Paassilta, M et al. | p-Poster | | | Miikka Paassilta[1], Rami Vainio[1], Angels Aran[2], Athanasios Papaioannou[3], Anastasios Anastasiadis[3], Piers Jiggens[4], Sigiava Giamini[5] | | | [1]University of Turku, Finland, [2]University of Barcelona, Spain, [3]National Observatory of Athens, Greece, [4]ESA/ESTEC, The Netherlands, [5] Space Applications & Research Consultancy (SPARC), Athens, Greece | | | Kahler & Ling (2017: Solar Physics 292:59) proposed a simple modified Weibull function to model the time–intensity profile of solar proton events and showed it to correspond well to the observations of about a dozen large energetic proton events. Motivated by this, using the solar energetic proton event list maintained by NOAA (Solar Proton Events Affecting the Earth Environment; available at ftp://ftp.swpc.noaa.gov/pub/indices/SPE.txt), we have fitted a modified Weibull function to SEPEM (GOES) 7.23–-10.46 MeV proton intensity data recorded during the listed events (spanning the years from 1976 to 2015, 217 events in total). For this purpose, the intensity data were smoothed with a 30-minute average and normalized so that event peak intensity equals 1.0 (which is also the maximum of the analytic profile). We find that in some 60$%$ of the considered events, the function models the behaviour of the proton intensity for a period of $>$24 hours following the event onset at least reasonably faithfully; the remaining cases result in failure primarily due to rapidly following new events (24$%$), data gaps (3$%$), elevated background due to previous events (4$%$), or a combination of these reasons. The fit function as applied by us has three free parameters, $\alpha$, $\beta$, and $t_0$; $\alpha$ controls the shape of the initial flux increase, $\beta$ controls the duration of the event and $t_0$ the time delay of the event onset from the flare. Of these, $\alpha$ appears to be fairly well correlated (through a quadratic dependence) with the solar longitude of the event-related flare, while this correlation is absent for the other parameters. Especially the events originating from the eastern hemisphere of the Sun show a considerable amount of variety in the shapes of their time profiles (rise and decay times), and work is underway to infer their relation to other parameters of the related solar activity than the flare longitude. We have also applied the fitting procedure to the flux profiles produced by the SOLPENCO2 tool. Results comparable to full set of observed events have been obtained indicating that SOLPENCO2 proton event model is based on a representative set of SEP events, thus providing further observational support for the model. We will discuss these results in proton time–intensity profile predictions in the framework of the ESA/SAWS/ASPECS project.
This work was supported by the European Space Administration (ESA) contract No. 4000120480/17/NL/LF/hh, "Solar Energetic Particle (SEP) Advanced Warning System (SAWS)". | | 6 | Processing Solar Images to forecast Coronal Mass Ejections using Artificial Intelligence | Savvas, R et al. | p-Poster | | | Savvas Raptis[1], Jorge Amaya[1], Adam Shamash[1], GIlles Depypere[1], Giovanni Lapenta[1] | | | [1] Centre for Mathematical Plasma-Astrophysics, KU Leuven, Leuven, Belgium | | | In this project, we develop a new processing tool that provides the input, which is used by Artificial Intelligence algorithms to create a new type of forecasting model for Coronal Mass Ejections. The reason A.I. models are used is because compared to atmospheric weather forecasting that we can rely on a large network of weather stations. In space, there are not enough satellites and a physics based modeling of interplanetary Space is limited.
The processing tool is a module, built in python, that downloads process and organize images from SDO in order to be fully utilized as inputs in the A.I. model that is being built. The idea behind the processing tool is to create an automatic procedure that uses Sun’s differential rotation and images from SDO to derive “History Maps”. These maps show how the intensity of a specific area of the Sun developed during a specific time period. The tool takes the intensity of multiple longitudes at a given date, and then tracks their position back in time ([x] hours), creating one history image per longitude. A 3D representation of the line’s intensity ([longitude vs latitude vs time]) keeps a recollection of the evolution of fast-evolving features in the atmosphere of the Sun, like CMEs. Multiple wavelengths can be used to also take into account the feature evolution in altitude.
The goal of the project is to forecast possible development of Coronal Mass Ejections (CMEs) from different part of the Sun along with their characteristic. CMEs are one of the energetic events that can lead to disturbances in the magnetic field of Earth and therefore are directly connected to Space Weather Applications. For the prediction model, Convolution Neural Networks (CNN) are used to obtain information from the CME catalog of CACTus and LASCO, using images from SDO. The input of our model is, therefore, a pre-processed time-series of 2D images from the Sun, whereas, the final output is the forecasting of a CME and its characteristics. Feature selection of our model is carefully done from a wide range of observations, that include different wavelengths, magnetometers etc. until the perfect combination of data is found. For the training of the CNN and for the image processing, we will be implementing HPC techniques using the Flemish Supercomputer Centre (VSC).
This work is part of the AIDA H2020 (www.aida-space.eu) project funded by the European Commission (EC).
| | 7 | Super-Resolution of Solar Images using Generative Adversarial Networks | Shamash, A | p-Poster | | | Many modeling and prediction tools require the use of solar images, creating a desire for high resolution imagery of the sun. Unfortunately, due to either technological or financial constraints, only low resolution images are available for use.
In this work, we propose a two-stage procedure for the super resolution of solar images. In other words, we seek a mapping from a low dimensional image of the sun to a high dimensional version of the same image. The first stage consists of a Generative Adversarial Network (GAN) designed to learn the wavelet coefficients of our high dimensional image from the low dimensional input image. In the second stage, we use a second GAN to learn the high dimensional image from these wavelet coefficients and the low dimensional. The architecture of this second network mimics the steps for an inverse wavelet transform, thus yielding a natural performance metric by comparing the terms in the true transformation with those found by our network. The combination of these two GANs thus provides a very natural way of super resolution that we expect will provide high quality solar images.
This work is part of the AIDA H2020 (www.aida-space.eu) project funded by the European Community (EC) | | 8 | The Advanced Solar Particle Events Casting System (ASPECS) activity | Anastasiadis, A et al. | p-Poster | | | A. Anastasiadis[1], A. Aran[2], R. Vainio[3], I. Sandberg[4], M. Dierckxsens[5], P. Jiggens[6], A. Papaioannou[1], M.K. Georgoulis[7], E. Paouris[1], G. Balasis[1], O. Giannakis[1], G. Vasalos[1], M. Paassilta[3], S. Aminalragia-Giamini[4], A. Tsigkanos[4] | | | [1]Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing (IAASARS), National Observatory of Athens, I. Metaxa & Vas. Pavlou St., GR-15236, Penteli, Greece, [2]Department of Quantum Physics and Astrophysics, Institute of Cosmos Sciences (ICCUB) ,Universitat de Barcelona, 08028 Barcelona, Spain, [3]Department of Physics and Astronomy, University of Turku, 20014 Turku, Finland, [4]Space Applications & Research Consultancy (SPARC), Athens, Greece, [5]Royal Belgian Institute for Space Aeronomy (BIRA-IASB), Avenue Circulaire 3, 1180 Uccle, Belgium, [6]European Space Research and Technology Centre (ESTEC), Space Environment and Effects Section, Keperlaan 1, 2200AG Noordwijk, The Netherlands, [7]Research Center of Astronomy and Applied Mathematics (RCAAM), Academy of Athens, Athens, Greece | | | Solar Energetic Particle (SEP) events can adversely affect space and ground-based systems. The space weather effects associated with SEP events include communication and navigation systems, spacecraft electronics and operations, space power systems, manned space missions, and commercial aircraft operations. As a result, there is a clear need for the development of a scheme which provides advanced warning and forecast of SEP events, their characteristics and time profiles. In this work, a new web based service for the prediction of solar eruptive and energetic particle events is presented. The ASPECS (Advanced Solar Particle Event Casting System) activity is designed to advance the technology development of a Solar Particle Radiation Advanced Warning System (SAWS). It will collate and combine outputs from different modules providing forecasts of solar phenomena, solar proton event occurrence and solar proton flux and duration characteristics; tailored to the needs of different spacecraft and launch operators, as well as the aviation sector. The predictions will start with the solar flare forecasting and continuously evolve through updates based on near-real time inputs (e.g. solar flare and coronal mass ejections data/characteristics) received by the system. User requirements include a derivation of energies and thresholds important for different user-groups and warning levels. A thorough validation process will provide results on the system’s forecasting accuracy and will be used to tailor the modules and the use of different data sources and algorithms for different forecast horizons and SEP energies.
This work was supported through the ESA Contract No 4000120480/17/NL/LF/hh "Solar Energetic Particle (SEP) Advanced Warning System (SAWS)"
| | 9 | Application of the SEPEM statistical modelling tool to the Helios 1 and 2 missions | Aran, A et al. | p-Poster | | | Angels Aran[1], Robert Florido[1], Piers Jiggens[2], Daniel Pacheco[1], Daniel Heynderickx[3], Blai Sanahuja[1] | | | [1] Dept. de Física Quàntica i Astrofísica, Institut de Ciències del Cosmos (ICCUB), Universitat de Barcelona, Spain [2] European Space Research and Technology Centre, ESA, The Netherlands [3] DH Consultancy, Leuven, Belgium | | | SEP events pose a serious hazard for both spacecraft and humans in interplanetary space. In the frame of the Solar Energetic Particle (SEP) Environment Modelling (SEPEM) Project of the European Space Agency (ESA), a statistical model to predict the proton radiation environment of interplanetary missions was developed. In this work, we use this unique tool, the ESA’s SEPEM Away from 1 AU tool (http://sepem.eu), for statistical modelling of the cumulative and the worst case proton fluences over the Helios 1 and Helios 2 missions. We compute their energy spectra for periods of the solar cycle over the mission duration and fit them to most common functions. We analyse the best fitting function for each case. Finally, we compare the SEP cumulative fluence energy spectra of the model results at a 90% confidence level with the fluence measured by the Helios 1 and 2 spacecraft over their entire mission durations. | | 11 | Pre-flare dynamics in 3D of the Active Regions | Korsos, M et al. | p-Poster | | | Marianna Korsos | | | ELTE | | | We introduce a novel pre-flare tracking of delta-type sunspot towards improving the estimation of flare onset time by focusing on the evolution of the 3D magnetic field construction of flaring ARs. The 3D magnetic structure is based on PF/NLFFF extrapolations and PENCIL MHD code simulations to encompassing a vertical range from the photosphere through the chromosphere and transition region into the low corona. The basis of our proxy measure of activity prediction is the so-called weighted horizontal gradient of magnetic field (WG_M) defined between umbrae of opposite polarities in the entire delta-type sunspot. The temporal variation of the distance of the barycenter of the opposite polarities is also found to possess potentially important diagnostic information about the flare onset time estimation as function of height. We found that at a certain height in the lower solar atmosphere the onset time may be estimated much earlier than at the photosphere or at any other heights. Therefore, we present a tool and recipe that may potentially identify the optimum height for flare prognostic in the solar atmosphere allowing to improve our flare prediction capability and capacity. This tool would be an excellent example how to exploit high-resolution ground-based observations with our future facilities, like DKISt or EST.
| | 12 | Long-term electron radiation belt data assimilation relying on four spacecraft, the VERB code, and a sequential Kalman filter | Cervantes villa, J et al. | p-Poster | | | Sebastián Cervantes[1,2], Yuri Shprits[1,2,3], Alexander Drozdov[3], Adam Kellerman[3] and Nikita Aseev[1,2] | | | [1] GFZ, German Research Centre for Geosciences, Potsdam, Germany, [2] University of Potsdam, Potsdam, Germany, [3] University of California, Los Angeles, CA, USA | | | The dynamical evolution of the radiation belts has been a subject of extensive research since their discovery in 1959. After decades of study, it is now known that they experience significant changes due to acceleration, loss and transport of particles trapped in the Earth’s magnetic field. Since high-energy electrons can potentially cause spacecraft anomalies and damage satellite hardware, understanding and predicting fluxes in the radiation belts is of great importance to satellite operators, engineers, and designers.
Nevertheless, analysis of radiation belt observations presents major challenges. Satellite observations are often restricted to a limited range of L-shells, pitch angles, and energies. Additionally, observations at different L-shells are taken at different points along the spacecraft orbit and therefore at different times. Moreover, particle fluxes vary on short time scales, and observations from a single spacecraft do not allow for measuring the temporal variations on time scales shorter than the spacecraft orbital period. Analysis is further complicated by the fact that measurements are contaminated by errors, which are different for various instruments. As a consequence, to fill the spatiotemporal gaps and to understand the dynamics and dominant physical processes in the radiation belts as well as to create accurate models, observations should be combined with physics based dynamical models in an optimal way.
A methodology, usually referred as data assimilation, in which observations are blended with models in order to produce results close to the true values, has been successfully applied in other fields, such as numerical weather prediction. In this study we implement a data assimilation tool using a split-operator sequential Kalman filter approach. Reanalysis of the electron radiation belt fluxes is obtained over the period October 2012 to October 2015 by combining sparse observations from the Van Allen Probes spacecraft and the Geostationary Operational Environmental Satellites 13 and 15 with the 3D Versatile Electron Radiation Belt code. At first, radial profiles of electron fluxes are reconstructed, and the innovation vector is analyzed to show how the data is correcting for physical mechanisms absent in the model. Such processes (mixed pitch angle-energy diffusion, losses to the magnetopause, and scattering by Electromagnetic Ion Cyclotron waves) are then added in the reanalysis, and a validation against LANL GPS data is presented. Finally, major improvements with respect to the pure physics based model are discussed. It is demonstrated that the 3D data assimilative code provides a comprehensive picture of the radiation belts and is an important step toward performing reanalysis using observations from current and future missions.
| | 13 | Anomalies of solar activity and phase synchronization of solar magnetic field. | Blanter, E et al. | p-Poster | | | Elena Blanter[1,2], Jean-Louis Le Mouël[3], Alexander Shapoval[1], Mikhail Shnirman[1,2], Vincent Courtillot[3] | | | [1]Institute of Earthquake Prediction Theory, Russian Academy of Sciences, [2]National Research University, Higher School of Economics, [3]Institute de Physique du Globe, Paris | | | Solar dynamo presents a natural example of a complex non-linear system, whose evolution combines quasi-regular oscillations (e.g. the solar cycle) with extreme events (critical changes, atypical cycles, Great minima etc). Despite the recent quick progress of the solar dynamo modeling, the extreme events and irregularities of solar dynamics are still poorly understood due to variability of their properties. Although each specific kind of solar extremes appears rarely, the atypical solar cycles are quite common and the understanding of their origin and features is essential for the solar activity and space weather forecasting.
We compare significant anomalies in the long-term evolution of solar activity with the evolution of meridional circulation speed reconstructed in the two solar hemispheres through a Kuramoto model of non-linear coupled oscillators. We show that the most prominent anomalies observed in the solar activity and related indices correspond to the breaks of phase synchronization.
The Kuramoto model (Acebron et al, 2005) describes phase synchronization in the oscillatory systems and we apply it to phase synchronization between the two components of solar magnetic field (toroidal and poloidal) considering the meridional circulation cells in two solar hemispheres as coupled oscillators. Solving the inverse problem we are able to reconstruct either the evolution of the natural frequencies of the meridional circulation under assumption of constant coupling or the evolution of coupling under assumption of constant speed of the meridional circulation. Breaks of the phase synchronization are manifested in the singularities of the reconstructed coupling or natural frequencies of the meridional circulation.
We suggest that the phase synchronization deserves further investigation as a new indicator of anomalies in the solar dynamo evolution.
E. Blanter, A. Shapoval, and M. Shnirman acknowledge the support of the Russian Science Foundation through project No 17-11-01052. | | 14 | On the use of topside RO derived electron density for model validation | Shaikh, M et al. | p-Poster | | | Muhammad Mubasshir Shaikh[1,4], Bruno Nava[2], Haris Haralambous[3] | | | [1] Department of Applied Physics and Astronomy, University of Sharjah; [2] T/ICT4D Laboratory, International Center for Theoretical Physics; [3] Department of Electrical Engineering, Frederick University, Nicosia, Cyprus; [4] Ionospheric Laboratory, Sharjah Center for Astronomy and Space Sciences | | | In this work, the standard Abel inversion has been exploited as a powerful observation tool which may be helpful to model the topside of the ionosphere and therefore to validate ionospheric models. A thorough investigation on the behavior of Radio Occultation (RO) derived topside electron density (Ne(h))-profiles has therefore been performed with the main purpose to understand whether it is possible to predict the accuracy of a single RO-retrieved topside by comparing the peak density and height of the retrieved profile to the true values. As a first step, a simulation study based on the use of the NeQuick2 model has been performed to show that, when the RO derived electron density peak and height match the true peak values, the full topside Ne(h)-profile may be considered accurate. In order to validate this hypothesis with experimental data, electron density profiles obtained from four different incoherent scatter radars have therefore been considered together with co-located RO-derived Ne(h)-profile. The evidence presented in this paper show that in all cases examined, if the ISR and the corresponding co-located RO profile have matching peak parameter values, their topsides are in very good agreement. The simulation results presented in this work also highlighted the importance of considering the occultation plane azimuth while inverting RO data to obtain Ne(h)-profile. In particular, they have indicated that there is a preferred range of azimuths of the occultation plane (80o – 100o) for which the difference between the "true" and the RO-retrieved Ne(h)-profile in the topside is generally minimal. | | 15 | Predicting and now-casting Kp index using historical and real-time observations | Shprits, Y et al. | p-Poster | | | Yuri Shprits[1], Ruggero Vasile[1], David Jackson[2], Claudia Stolle[1], Zhelavskaya Ira[1], Sean Bruinsma[3 ] | | | [1] GFZ, [2] Met Office, [3] CNES, | | | Previous studies have shown that solar wind based prediction of the Kp index can provide accurate short term forecast. This forecast can be further improved by using recent measurements and recurrence in Kp or recurrence in the solar wind. However, how much each of these methods can contribute to the prediction and how do predictions depend on the horizon time remains relatively unexplored. Understanding of the efficiently and limitations of the predictions based on each of the input data sets can help develop independent algorithms for predictions. Various combinations of such models that may depend on the availability of data may be used to produce a short term or long term forecast. Systematic analysis may also help identify the limits of predictability and how stochastic processes may contribute to the dynamics of the Kp index.
In this study we use the same set up of the codes with second order scheme for FFNN to examine how models using various input data can contribute to the accuracy of the long and short term predictions. The results are compared and the prediction RMSE is used to determine for which time horizons solar wind , persistence or recurrence based predictions provide most accurate results. We also study how efficient these models are during the disturbed conditions and perform re-normalization of data to improve the efficiency of the algorithms. This research is supported by H2020 SWAMI project. | | 16 | Coronal holes detection using supervised classification | Delouille, V et al. | p-Poster | | | Veronique Delouille[1], Stefan Hofmeister[2], Martin Reiss[3], Benjamin Mampaey[1], Manuela Temmer[2] | | | [1]Royal Observatory of Belgium, Belgium, [2]University of Graz, Austria, [3]Space Research Institute, Graz, Austria | | | We demonstrate the use of machine learning algorithms in combination with segmentation techniques in order to distinguish coronal holes and filaments in solar EUV images. We used the Spatial Possibilistic Clustering Algorithm (SPoCA) to prepare data sets of manually labeled coronal hole and filament channel regions present on the Sun during the time range 2010-2016. By mapping the extracted regions from EUV observations onto HMI line-of-sight magnetograms we also include their magnetic characteristics. We computed averagelatitude, area, shape measures from the segmented binary maps as well as first order, and second order texture statistics from the segmented regions in the EUV images and magnetograms. These attributes were used for data mining investigations to identify the most performant rule to differentiate between coronal holes
and filament channels, taking into account the imbalance in our dataset which contains one filament channel for 15 coronal holes. We tested classifiers such as Support Vector Machine, Linear Support Vector Machine, Decision Tree, k-Nearest Neighbors, as well as ensemble classifier based on Decision Trees. Best performance in terms of True Skill Statistics are obtained with cost-sensitive learning, Support Vector Machine classifiers, and when HMI attributes are included in the dataset. | | 17 | Forecast of fast solar wind using global 3D MHD simulation from the Sun to 1AU with an empirical coronal heating model | Den, M et al. | p-Poster | | | Mitsue Den[1], Takashi Tanaka[1,2], Yuki Kubo[1], Shinichi Watari[1] | | | [1]National Institute of Information and Communications Technology, [2]Kyushu University | | | Fast solar wind forms corotating interaction region (CIR) with slow solar wind and this causes recurrent
geomagnetic disturbances in the Earth’s magnetosphere. The source of fast solar wind is coronal holes (CHs),
which exist both during solar minimum and maximum. Fast solar wind arrives in every solar rotation, so
prediction of occurrence and timing of fast solar wind and formed CIR is one of the important issues in
space weather forecasting. For evaluation of the effect of a coronal hole on Earth’s magnetospheric
activity, it is still difficult to estimate the case that CHs extend from the high latitude to the low
latitude. Using our three-dimensional MHD simulation code, REPPU (REProduce Plasma Universe) code driven
by solar magnetic field from the sun to 1AU, we are monitoring the global solar wind condition since 2016.
Our model is able to reproduce both coronal holes and global solar wind structure, so it can connect fast
solar wind observed at L1 point and that origin. We present our simulation results and discuss the effect
of an empirical coronal heating model and solar wind acceleration region on reproduce of fast solar wind.
| | 18 | Solar Predict: A 4-D var method for hind- and forecasting solar activity | Brun, A et al. | p-Poster | | | Allan Sacha Brun, Ching Pui Hung, Alexandre Fournier, Laurène Jouve, Antoine Strugarek, Olivier Talagrand | | | [1] DAp-AIM, [2] DAp-AIM, [3] IPGP, [4] IRAP, [5] DAp-AIM, [6] LMD | | | We have developed a 4D Var data assimilation method to reconstruct past and future solar activity. We first test its performance by inferring the time varying and multi-cellular internal meridional circulation used to create synthetic magnetic proxy data via a mean field Babcock-Leighton dynamo model. We then apply it over 40 years of Wilcox Observatory solar magnetograms , hence inferring the solar meridional circulation over the last 3 sunspot cycles. We then use the data assimilation pipeline to make a first forecast of cycle 25. | | 19 | MUF(3000)F2 maps over Europe by Kriging interpolation method | Sabbagh, D et al. | p-Poster | | | Dario Sabbagh[1], Carlo Scotto[1] | | | [1] Istituto Nazionale di Geofisica e Vulcanologia, Via di Vigna Murata 605, 00143 Rome, Italy | | | A method for producing MUF(3000)F2 maps over Europe upgrading background maps have been developed and tested in different conditions. The value of the variable z is interpolated by the Ordinary Kriging method, being z=(MUF(3000)F2[measured]-MUF(3000)F2[median])/MUF(3000)F2[median] for a given time, where MUF(3000)F2[median] is the value obtained by the IRI-CCIR model. The interpolation is performed in the region of longitude 20°W-30°E and latitude 32°N-65°N, over a grid with spatial resolution 0.5°x0.5°, and using measured values from different available ionospheric stations. Then, upgraded MUF(3000)F2 maps are produced over the interpolation grid as MUF(3000)F2=(1+z)*MUF(3000)F2[median].
MUF(3000)F2 values recorded both in quiet and disturbed geomagnetic conditions have been used to test the method and choose the best variogram model for the Kriging procedure. For each time considered, the interpolation has been performed using five variogram models (Gaussian, spherical, exponential, power and linear), and results have been compared on the basis of the usual statistical tests on the variogram models and the RMSE values obtained at a test station between modeled and measured MUF(3000)F2 values.
This method is meant to be used for the European MUF(3000)F2 nowcasting service offered by the Pan-European Consortium for Aviation Space Weather User Services (PECASUS), when real-time data recorded every 15 minutes will be used for the automatic maps production and dissemination. | | 20 | A method for MUF(3000) short-term (1-24) hour prediction over Europe. | Perrone, L et al. | p-Poster | | | L. Perrone[1] and A.V. Mikhailov[2] | | | [1]Istituto Nazionale di Geofisica e Vulcanologia (INGV); [2]Pushkov Institute of Terrestrial Magnetism, Ionosphere and Radio Wave Propagation (IZMIRAN) | | | A method for a short-term MUF(3000) forecasting (1-24 hours ahead) over European region has been proposed. MUF(3000) depends on two ionospheric parameters foF2 and M(3000):
foF2 is predicted the EUROMAP model and for M(3000) is taken from the IRI.
There is a need to develop methods for the ionospheric F2-layer short-term (1-24) hour prediction. Various approaches can be found in literature. Today the priority should be given to empirical approaches as they provide higher prediction accuracy compared to physical (first-principle) models. A method which provides acceptable results under various geophysical conditions and which can be used in practice has been considered. The method has been applied to Europe where there are ionospheric stations with long (for some solar cycles) historical data and current real-time foF2 observations. A prediction method should be based on Ap indices as only daily mean Ap is predicted at present with 1-3 day lead time, and this is a serious limitation.
The method includes two types of prediction models: regression models based on the analyses of historical observations, and training models based on current foF2 observations.
A mapping procedure applied to the European stations provides MUF(3000) short-term prediction over the whole area. This model is inserted in Pan-European Consortium for Aviation Space weather User Services (PECASUS).
| | 21 | Innovative methods for the prediction of solar flares: data assimilation in sandpile models with machine learning | Strugarek, A et al. | p-Poster | | | Antoine Strugarek[1], Benoît Tremblay[2], Paul Charbonneau[2], Allan Sacha Brun[1], Nicole Vilmer[3] | | | [1] Laboratoire AIM Paris-Saclay, CEA/Irfu, Gif-sur-Yvette, France, [2]Département de physique, Université de Montréal, Montréal, Canada, [3]LESIA, Observatoire Paris, Meudon, France | | | The largest solar flares, of class X and beyond, are often associated with strong energetic particles acceleration. The acceleration process scenario based on multiple reconnection sites can be modelled with so-called cellular automata or sandpile models. These toy models have the advantage of reproducing many statistical features of solar flares, but generally lack a sound physical justification.
Building on the pioneering work of Lu and Hamilton, we develop a new class of sandpile models motivated by a physical interpretation of the three main ingredients (driver, instability criterion, relaxation path to a stable state) of self-organized criticality models. These models reproduce satisfyingly the statistical properties of solar flares, and we show that they possess very promising predictive capabilities for the largest events, in spite of their embedded stochastic process. We report on an ongoing effort in coupling the sandpile models with data assimilation techniques using the GOES X-ray flux, leveraging the power of machine learning to accelerate the assimilation process, and give preliminary estimates of its predictive performance using the past largest solar flares.
| | 22 | Precise estimation of the delayed ionospheric response to solar EUV variations | Schmölter, E et al. | p-Poster | | | Erik Schmölter[1], Jens Berdermann[1], Norbert Jakowski[1], Christoph Jacobi[2], Rajesh Vaishnav[2] | | | [1]German Aerospace Center, [2]University Leipzig, Institute for Meteorology | | | The solar EUV radiation has a strong influence on the ionosphere and interacts with the ionospheric plasma in various ways through physical and chemical processes. These interactions cause a response of the ionosphere against EUV variations with a delay of roughly one day. A precise estimation and a better understanding of this delay is important for the validation and improvement of existing ionospheric models. This will improve the forecast of the ionospheric conditions and allows to mitigate its effects on GNSS services. We present our analysis of the ionospheric delay with focus on seasonal, diurnal and geographic dependencies and will discuss our findings in connection with the ionospheric processes involved. Furthermore, we will demonstrate why physical modelling is needed to find a consistent explanation of the given results. | | 23 | Forecasting the Strength of Geomagnetic Storms utilizing CME-ICME characteristics | Paouris, E et al. | p-Poster | | | Evangelos Paouris, Athanasios Papaioannou, Anastasios Anastasiadis, Georgios Balasis | | | Institute for Astronomy, Astrophysics, Space Applications & Remote Sensing National Observatory of Athens, Greece | | | Minor and moderate geomagnetic storms are triggered by the arrival of coronal mass ejections (CMEs) and high-speed streams (HSS) of solar wind while strong and severe geomagnetic storms are triggered by fast and wide CMEs. In this work the prediction of the strength of the geomagnetic storms, as this is introduced by NOAA scales (i.e. from G1 to G5), based on the characteristics of the parent CME-ICME associated events, is presented. In particular, we employ the linear speed (VCME) and the angular width (w) of the CME, the average and maximum speed (V and Vmax) as well as the magnetic field (B) and its zth-component (Bz), inside the sheath or the ICME. An optimal combination of these variables is established, leading first to an estimation of the possible maximum value of the geomagnetic index Ap and, from there to the level (G-class) of the potential geomagnetic storm.
This work was supported by the project “PROTEAS II” (MIS 5002515), which is implemented under the Action “Reinforcement of the Research and Innovation Infrastructure”, funded by the Operational Programme “Competitiveness, Entrepreneur- ship and Innovation” (NSRF 2014–2020) and co-financed by Greece and the European Union (European Regional Development Fund). | | 24 | The dependence of high-speed stream peak velocities and of the Kp index on the positions of their source coronal holes on the sun | Hofmeister, S et al. | p-Poster | | | Stefan Hofmeister[1], Astrid Veronig[1], Manuela Temmer[1], Susanne Vennerstrom[2], Bernd Heber[3], Bojan Vršnak[4] | | | [1] Institute of Physics, University of Graz, Graz, Austria, [2] DTU Space, National Space Institute, Kongens Lyngby, Denmark, [3] Institut für Experimentelle und Angewandte Physik, Universität Kiel, Kiel, Germany, [4] Hvar Observatory, Faculty of Geodesy, Zagreb, Croatia | | | The peak velocities of high-speed solar wind streams are linearly dependent on the areas of their solar source regions, coronal holes (velocity-area relation). We revise this relationship in a comprehensive study of 115 events in the time-range from 2010/08 to 2017/03, and analyze the properties of the solar source coronal holes, the corresponding high-speed solar wind streams at 1 AU, and for a subset their geomagnetic consequence, using remote-sensing and in-situ data from the satellites SDO, ACE, STEREO-A, STEREO-B, and the Kp index.
We find a further distinct dependence of the high-speed stream peak velocities as measured by the satellites and of the Kp index on the co-latitudes of the solar source coronal holes, that is the latitudinal angle between the coronal hole on the Sun and the in-situ measuring satellite. High-speed streams arising from coronal holes located near the ecliptic result in the highest peak velocities and Kp index per coronal hole area, and they linearly decrease with increasing co-latitudes of the source coronal holes up to co-latitudes of ~60°. By adding the coronal hole co-latitudes as a further parameter to the velocity-area relationship, the Spearman‘s correlation coefficient between predicted and measured high-speed stream peak velocities increases from cc=0.50 to cc=0.72.
We interpret this as an effect of the three-dimensional propagation of high-speed streams in the heliosphere: when the source coronal hole is located in the ecliptic directly facing the measuring satellite, then the satellite is passing the central part of the corresponding high-speed stream measuring the highest plasma velocities of the high-speed stream. Whereas,
when the source coronal hole is located at higher solar latitudes, the satellite is passing the latitudinal flank of the high-speed stream, measuring much lower plasma velocities.
We conclude that the latitudinal angle between the source coronal hole on the Sun and the measuring satellite is an estimate on the location of the satellite within the high-speed stream. Therefore, the co-latitude is an important parameter for the forecast of high-speed stream parameters near the Earth and their related geomagnetic storms. | | 25 | High-speed solar wind stream forecast based on coronal hole data | Podladchikova, T et al. | p-Poster | | | Tatiana Podladchikova[1], Astrid M. Veronig[2], Manuela Temmer[2], Stefan Hofmeister[2] | | | [1] Skolkovo Institute of Science and Technology, Russia, [2] Kanzelhöhe Observatory & Institute of Physics/IGAM, University of Graz, Austria | | | Accurate solar wind modeling is important for the prediction of geomagnetic response of high-speed solar wind streams as well as for modeling the transit of coronal mass ejections in interplanetary space and their impact at Earth. Data assimilation techniques combining the strength of models and observations provide a very useful tool for accurate solar wind forecasts. We develop a method to predict high-speed solar wind streams at Earth 1-day ahead by using coronal hole areas derived from SDO AIA images in combination with in situ solar wind plasma and field data (speed, density, and magnetic field magnitude) from the ACE and Wind spacecraft. Our approach integrates two types of forecast: on the basis of solar wind data, and coronal hole data. The forecast on the basis of solar wind data uses a multidimensional linear regression model relating the solar wind speed one day ahead with the proton density, magnetic field magnitude, and solar wind speed at the current moment. One of the major concerns with such data assimilation scheme is that the regression coefficients do not remain constant and are time-varying. To avoid the fitting of regression coefficients to a particular situation, that can vary in future, we develop a Kalman filter to create a dynamic linear regression for the 1-day ahead prediction of the solar wind speed. The basic idea behind the forecast on the basis of coronal hole data is to use the high correlation (0.7-0.9) found between the ascending and descending phases of coronal hole areas on the Sun and solar wind speed at Earth. The main uncertainty in this scheme is the time shift between coronal hole area observations and the following increase in solar wind speed, that varies between 1 – 5 days. We propose to estimate this shift with first signs of coronal hole and solar wind increase. As the peak in coronal hole area is mostly observed before the development of the peak in the solar wind, we predict also the time of peak in solar wind speed. The integration of the forecast on the basis of coronal hole data with the forecast on the basis of solar wind data is performed with an adaptive Kalman filter. The approach is tested for the period 2010-2017 and can be useful for the operational forecasting of the space weather conditions in the near-Earth environment.
| | 26 | Multi-spacecraft Prediction of Co-rotating Stream Interaction Regions | Vennerstrom, S et al. | p-Poster | | | Susanne Vennerstrom[1], Astrid M. Veronig[2], Manuela Temmer[2], Stefan Hofmeister[2] and Stephan G. Heinemann[2] | | | [1]Technical University of Denmark, DTU Space, [2]2 Institute for Physics, University of Graz, Austria | | | Prediction of the solar wind parameters near Earth with a lead time larger than 1-2 hours are currently one of the major challenges in forecasting geomagnetic activity and other space weather effects associated with solar wind – magnetosphere coupling.
The service AWARE (Automated WARnings of Earth arrivals) is currently operating at near real-time in situ data from L1 and STEREO A to provide early automated warnings of arriving ICMEs and CIRs. We present a new development of this service called AWARE_Next, which combine the in situ detections in L1 and STEREO A to provide a longer lead time forecast of arriving stream interaction regions (SIRs) and high-speed streams. The forecast is based on a statistical investigation of the physical characteristics and recurrence of SIRs in the last solar cycle. The forecast has a special focus on the „next 24 hours“.
The new service constitute part of the ESA SSA Heliospheric Expert Service Center service provision.
| | 27 | Inferring dynamic causal time lag: Applications to space weather | Chandorkar, M et al. | p-Poster | | | Mandar Chandorkar[1, 2], Enrico Camporeale[1], Cyril Furtlehner[2], Michélè Sebag[2] | | | [1]Centrum Wiskunde Informatica Amsterdam; [2]INRIA Saclay - Île-de-France | | | It is often the case with natural and man-made phenomena, that cause and effect are temporally separated i.e. there is a time lag between occurrence of an event and the observation of its consequences. In complex systems, this time lag between cause and effect can be uncertain and dynamic. Mathematically this can be expressed as $y_{t + g(x_{t})} = f(x_{t})$, where $x_{t}$ is a time series representing the causes, $y_{t}$ represents the effects and functions $f$ and $g$ represent the input-output and input-time lag relationships.
In the context of space weather, one can see this when events on the Sun such as coronal mass ejections (CME) or high speed solar wind streams cause disturbances in the Earth’s geomagnetic state, several hours, or even days later. To increase the prediction window of space weather forecasting systems, it is important to model the temporal relationship between space weather drivers and geomagnetic quantities in the vicinity of the Earth.
We present ongoing work in learning dynamic causal time lags from noisy time series. Our methodology is based on a neural network which learns the input-output and input-time lag relationships simultaneously. We evaluate our models performance on a set of toy problems as well as on the problem of CME propagation using SDO and OMNI data sets. | | 28 | Verification of space weather forecast models: administrative, economic, and scientific | Wintoft, P et al. | p-Poster | | | P. Wintoft, M. Wik | | | Swedish Institute of Space Physics | | | Any model that provides predictions must be verified. This can be a complex task even for scalar valued time series and the purpose of the verification must be specified. There are basically three categories of verification: administrative, economic, and scientific. There are some degree of overlap between the different categories, but the complexity can be reduced by identifying the purpose of the verification. Administrative verification is mainly concerned with the comparison between forecast models, whether it is different models or the improvement of models, and this usually requires a small set of performance measures. Thus, it is critical that the measures are appropriate and well defined. Economic verification includes the user requirements and tailored metrics, and different users may have different requirements even for the same underlying physical phenomena. Scientific verification is about understanding model performance that can lead to further improvements, but also understanding the limits and under which conditions the model fails. Scientific verification should thus not be limited to a few measures but instead apply many different strategies. Independent high quality extensive datasets are required, both considering model inputs and outputs. As forecast models based on machine learning (ML) techniques derive their functions (or mappings) from the data, and can in principle have unlimited number of free parameters, it is crucial to have an independent dataset for verification. However, their low computational demand sets usually no limits on producing extensive time series. We discuss the above and apply it to ML model predictions of Kp, Dst, and AE time series. | | 30 | A multiscale artificial neural network approach to geomagnetic index forecasting | Consolini, G et al. | p-Poster | | | G. Consolini[1], T. Alberti[1], F. Giannattasio[2] and P. De Michelis[2] | | | [1]INAF-Institute for Space Astrophysics and Planetology, Roma, Italy, [2]Istituto Nazionale di Geofisica e Vulcanologia, Roma, Italy | | | The Earth's magnetosphere presents a complex dynamical behavior in response to the changes of the solar wind and interplanetary medium conditions. In this framework, forecasting geomagnetic indices behavior in response to solar wind changes is a challenging problem. Indeed, it comprises both slow and fast externally driven and purely internal processes which are characterized by an inherent multiscale nature.
A quantitative evidence of a clear timescale separation between externally-driven and internal processes has been recently shown by Alberti et al. [1], so that a reliable forecast of the Earth's magnetospheric response would include the capability to correctly predict the processes occurring on faster timescale, which have a great impact in the framework of Space Weather.
The purpose of the current study is to discuss in detail the multiscale nature of the Earth's magnetosheric dynamics and to present an attempt to forecast the geomagnetic indices behavior using a Multiscale Artificial Neural Network (MANN) approach by considering the interplanetary magnetic field (IMF) measurements as a external driver. This approach requires the simultaneous use of methods to disentangle the multiscale character of the magnetospheric response and ad hoc scale-by-scale Artificial Neural Networks.
[1] Alberti, T., et al., Timescale separation in the solar wind-magnetosphere coupling during St. Patrick's Day storms in 2013 and 2015, \textit{J. Geophys. Res. Space Physics}, \textbf{122}, 4266, 2017, doi:10.1002/2016JA023175
| | 31 | ACE/EPAM Solar energetic electron catalog (1996-2017): Statistical Relationship with Flares and CMEs | Samwel, S et al. | p-Poster | | | Susan Samwel[1], Rositsa Miteva[2] | | | [1] National Research Institute of Astronomy and Geophysics (NRIAG), 1142, Helwan,1 Cairo, Egypt; [2] Space Research and Technology Institute – Bulgarian Academy of Sciences, 1113 Sofia, Bulgaria | | | A newly comprehensive solar energetic electron catalog is compiled. We used the data from ACE/EPAM in the two energy channels 103–175 KeV and 175–315 KeV over two solar cycles 23 and 24. We identify the onset times, peak intensities and peak times of the electron events in addition to the characteristic quantities of the associated solar flares and coronal mass ejection. Finally, we discuss the statistical trends of the electron events in terms of their correlation with their solar origins for the two solar cycles 23 &24. | | 32 | Operational Performance of the COMESEP Alert System | Dierckxsens, M et al. | p-Poster | | | Mark Dierckxsens[1], Norma Crosby[1], Stijn Calders[1], Lenka Zychová[1] | | | [1] Royal Belgian Institute for Space Aeronomy, Ukkel, Belgium | | | The FP7 COMESEP (COronal Mass Ejections and Solar Energetic Particles: forecasting the space weather impact) project developed tools for forecasting geomagnetic storms and solar energetic particle (SEP) radiation storms. Alerts are provided in the form of a risk level based on the combination of the likelihood of occurrence and expected impact. All the necessary information to evaluate this risk level is received through the COMESEP Alert System which also disseminates the alerts to the end-user. The system has been operational since November 2013 and was integrated into the ESA Space Situational Awareness space weather network in September 2016. The forecasts issued during the operational period are compared with the observed conditions to evaluate the performance of the current system. A selected set of missed events and false alarms will be described in more detail in order to identify potential improvements. The technical performance and specific issues encountered while running the operational system will also be discussed. | | 33 | Connecting the Sun to the Earth using Machine Learning | Amaya, J et al. | p-Poster | | | Jorge Amaya, Adam Shamash, Gilles Depypere, Savvas Raptis, Diego Gonzalez-Herrero, Giovanni Lapenta | | | CmPA, KU Leuven, Belgium | | | A good forecast of atmospheric weather can be obtained by combining supercomputers with data assimilation techniques based on in-situ measurements of the atmospheric conditions. To achieve similar levels of forecasting accuracy in Space Weather, it is imperative to perform in-situ measurements of the solar wind conditions at multiple points between the Sun and the Earth.
However, this is currently impossible. Constant satellite measurements of the solar wind are only taken 1.5M Km ahead of the Earth, leaving a terribly large blind spot of almost 150M Km between the Sun and the Earth. MHD computer simulations of the propagation of the solar wind and CMEs, based on remote observations of the Solar atmosphere, can not be constrained and can lead to very inaccurate forecasts.
An alternative to MHD computer models is the use of Machine Learning techniques. In this approach a computer algorithm goes through a large historical record of data and finds correlations between inputs and outputs. In this work we present the initial results of our efforts to connect the Sun to the Earth using Machine Learning techniques.
We divide the Sun-Earth system in four segments: a) processing of the remote sensing data, b) connecting the photosphere to the corona, c) the corona to the solar wind at the L1 orbit, and d) connecting L1 to the Earth. For each one of the systems we build and test an independent machine learning model. This work has been performed in the context of three master of science thesis at KU Leuven.
The machine learning models achieved the following objectives: solar image resolutions have been enhanced using Generative Adversarial Networks (GANs). SDO images were used to forecast the emergence of CMEs on the Sun; the computer automatically found correlations between solar images and CME catalogs using Convolutional Neural Networks (CNNs). Solar wind classes at the L1 orbit have been predicted from CME catalogs. Finally geomagnetic storms have been predicted with great accuracy using simple Multi-Layer Perceptron (MLP) models and data from OMNIweb. | | 34 | The Self-Adjusted Solar Flux Forecasting Tool (SASFF) | Podladchikova, E et al. | p-Poster | | | Olena Podladchikova, Christophe Marque, Koen Stegen, David Berghmans and SIDC forecasting team | | | Royal Observatory of Belgium | | | Measurements of the solar microwave flux at 10.7 cm by Pentincton Observatory, Canada are widely used as a solar proxy index for the modelling of the upper atmosphere. SIDC currently provides 3-day forecasts for the F10.7 index - based on a manual analysis of the index values during the latest three Carrington rotations. Additional considerations are taken into account when new active regions emerge or flare probability is elevated.
In this work, we present recent results on the development of a new autonomous tool, the Self-Adjusted Solar Flux Forecasting (SASFF) tool http://solwww.oma.be/users/olenapo/F10.7_realtime/). The solar radioflux dynamics is modelled by a nonstationary random walk with a variable drift and the forecast of the F10.7 index is based on an adaptive Kalman filter model which is used to identify the a priori unknown drift.
SASFF operates autonomously and provides solar flux forecasts, for the next 3 days, including forecasting uncertainties. In addition, it issues alerts when the F10.7 index values are elevated due to the appearance of solar activity events. The dynamics of the forecasting errors as a function of the solar cycle is discussed. It is unambiguously demonstrated the significant forecasting precision of SASFF with respect to existing forecasting approaches, using as example a number of extreme events.
SASFF tool will be calibrated using solar radioflux measurements of several radio-observatories and will be further optimized to output continuous forecasts of the F10.7 index values providing automatic corrections of the atmospheric drag coefficients that are used during satellite de-orbiting operations. | | 35 | Photospheric Magnetic Field Properties of Flaring vs. Flare-quiet active regions, V: Results from HMI | Leka, K et al. | p-Poster | | | KD Leka[1,2], G Barnes[1] | | | [1] NWRA, [2] Nagoya University / ISEE | | | What constitutes the difference between those solar active regions that produce energetic events and those that do not? The answer no doubt lies in the state and ongoing evolution of the magnetic field. Extending this series of studies of the photospheric magnetic field as related to flare imminency, we consider daily evaluations of almost all HMI Active Region Patches (HARPS), including temporal evolution. Using the NWRA Classification Infrastructure based on NonParametric Discriminant Analysis, we evaluate not only the static characterization of the photospheric field (extending well beyond the SHARP parameters) but include coronal topology and time-series considerations, as well. Additionally, we extend the analysis beyond "global" parametrizations to describe sub-area sites which may play roles in coronal energization and event triggering. We report here on those parametrizations which best distinguish imminent flaring from imminent quiet sunspot groups.
This material is based upon work supported by the National Science Foundation under Grant No. 1630454 and Nagoya University / ISEE. |
|
|
