## Session 17 - Data Assimilation for Space Weather Applications

Matthew Lang (LSCE), Sacha Brun (CEA-Saclay)
Friday 22/11, 14:00-15:15
Elisabeth

Data Assimilation (DA) is the systematic combination of observations and model information to provide the best possible model evolution and its uncertainty estimate. DA is able to do this by taking into account the uncertainties present in observations, forecasts and the model itself, in order to minimise errors within the model simulations. It is an essential tool for weather forecasting, providing optimal starting points that reduce the impacts of the 'butterfly effect' in forecasts.
Data assimilation has long been under-developed in space weather applications. However, this is currently changing and experiments using data assimilation within the space weather community have been performed with encouraging results. Space Weather Data Assimilation (SWDA) is currently comprised of five main areas, which are:
1) Coronal data assimilation, where DA is used to assimilate observations of the Sun’s corona to improve initial conditions for solar wind modelling and understanding of the Corona
2) Solar wind data assimilation, which is the assimilation of in-situ and remote observations of the solar wind to improve forecasts/understanding of the solar wind;
3) The assimilation of observations during extreme solar events, to aid the forecasting of potentially damaging solar wind events, such as Coronal Mass Ejections or Solar flares;
4) Ionospheric data assimilation, which aims to infer properties of the Ionosphere by assimilating observations of the solar wind and geomagnetic field, important for improvements in GPS;
5) Solar-/Geo-magnetic field data assimilation, which uses observations of magnetic field (e.g from magnetograms) to infer properties/improve dynamo models of the Sun/Earth.

These fields of data assimilation all face different problems that are not encountered within meteorological data assimilation, such as the inclusion of magnetic fields, supersonic solar wind conditions and poorly understood generation mechanisms. Whilst each field aims to solve different problems, they are intricately linked, however, there is currently very little collaboration between these fields.

Talks
Friday November 22, 14:00 - 15:15, Elisabeth

 14:00 Data driving and data assimilation in EUHFORIA Poedts, S et al. Invited Oral Stefaan Poedts KU Leuven EUHFORIA (European heliospheric forecasting information asset) is a state-of-the-art MHD 3D time-dependent MHD model for the heliospheric wind and the evolution of superimposed (cone and flux rope) CMEs from 0.1 AU to 2.1 AU. The results feature a highly dynamic heliosphere that the model is able to capture in good agreement with in situ observations. EUHFORIA is a data-driven model and its coronal part is based on the semi-empirical Wang-Sheeley-Arge (WSA) model that uses GONG magnetograms (Janus maps) as boundary conditions for the potential field source surface (PFSS) extrapolations1 of the coronal magnetic field. The model was recently extended to also enable the use of MWO, MDI and HMI CR synoptic magnetograms. The WSA model derives the solar wind speed from the resulting 3D coronal field. The MHD wind solution uses this speed as a boundary condition at 0.1AU and stretches out to 2.1 AU. Also the CME parameters use observational data, namely coronagraph imagery. Indeed, the CME speed, insertion time, longitude, latitude and half width of the CMEs are obtained from 3D reconstruction techniques with the graduated cylindrical shell (GCS) model developed by Thernisien et al. In addition, for the flux-rope CME models, the tilt, helicity and the toroidal magnetic flux are determined from magnetic and EUV observations. Attempts are being made to improve the heliospheric part of EUHFORIA by applying machine learning and artificial intelligence methods for the data assimilation of solar wind information and the selection of CME initialization parameters. This is tough because whatever happens in the expanding solar wind has zero effect on the Sun: measuring a quantity at the Earth's orbit does not propagate any improvement on the model closer to the Sun, any improvement on data assimilation propagates only anti sunward. Given these difficulties, we have identified a technique that can address these complexities: the representer technique (RT) [Skandrani et al, 2014], also used in ocean modelling. The RT is based on a mathematical reformulation of the Kalman gain matrix in terms of representers and modulation factors that express the domain of influence that a measurement has on improving a forecast model, i.e. it expresses the spatial extent and the improvement factor on the forecast that is obtained by assimilating a given measurement at a given point. We are trying to apply this technique to the EUHFORIA models of the solar wind and of expanding CMEs to investigate the effect of simulating data from existing and virtual future missions. An advantage of the RT is that it provides also the ability of analyzing what improvement would be gain by data one does not actually have: this is a feature of great value for planning future, possibly multispacecraft, missions. 14:15 Development of adaptive Kalman filter for short-term forecasts of the F30 and F10.7 cm radio flux Podladchikova, T et al. Oral Tatiana Podladchikova[1], Olena Podladchikova[2], Astrid M. Veronig[3,4] [1] Skolkovo Institute of Science and Technology, Moscow, Russia, [2] Solar-Terrestrial Centre of Excellence, Royal Observatory of Belgium, [3] Institute of Physics, University of Graz, Austria, [4] Kanzelhöhe Observatory of Solar and Environmental Research, University of Graz, Austria Solar activity indices such as the F30 and F10.7 cm radio flux are required by most models characterizing the state of the upper Earth atmosphere, such as the thermosphere and ionosphere, in order to specify satellite orbits, re-entry services, collision avoidance maneuvers and modeling of space debris evolution. With the aim of advancing current forecasting capabilities, we develop a novel prediction method of the F30 and F10.7 solar indices 1 to 3 day-ahead using an adaptive Kalman filter. Traditional approaches to the solar radio flux forecasting are based on linear regression models. However, one of the major concerns with such data assimilation scheme is that the evolution of the solar radio flux is a strongly non-stationary process, and thus the use of constant regression coefficients cannot be an optimal selection. The Kalman filter technique removes this disadvantage by adjusting the regression coefficients in real-time during the observation period, thereby increasing the forecasting accuracy. Testing the developed prediction technique over the period 2004-2016, we obtain a correlation coefficient between the predicted and observed values of about 0.99 (1-day ahead) and 0.98 (3-day ahead) for the F30 index and 0.99 (1-day ahead) and 0.96 (3-day ahead) for the F10.7 index. We compared our forecasts with the results provided by the few currently operating models for various periods of the solar cycle (minimum, ascending, maximum and declining phases). The RMS errors of the predictions are reduced by 3-14% in comparison with the predictions of F30 index based on neural networks (CLS) and the F10.7 index provided by the SIDC manual analysis. Additionally, the developed technique produces non-biased predictions, which is a strong advantage compared to the CLS forecast (overestimation) and SIDC forecast (underestimation). Thus, the proposed adaptive Kalman filter method significantly improves the quality of the F30 and F10.7 cm solar radio flux predictions and can be recommended for space weather applications. 14:30 Nowcasting the Ionosphere-Thermosphere System during Disturbed Conditions Sutton, E et al. Oral Eric Sutton[1], Jeffrey Thayer[1], Thomas Berger[1], Marcin Pilinski[2] [1] Space Weather Technology, Research, and Education Center (SWx TREC), University of Colorado [2] Laboratory for Atmospheric and Space Physics (LASP), University of Colorado The ionosphere-thermosphere (I-T) system rarely reaches a quiescent state, making prediction a persistent, day-to-day effort. In addition to advance knowledge of upstream space weather activity, accurate modeling and data assimilation capabilities are required to establish the initial conditions of any global-scale I-T forecast. Using Iterative Re-Initialization, Driver Estimation and Assimilation (IRIDEA), a recently developed approach for estimating real-time corrections to external drivers (e.g., solar wind and EUV irradiance variability) and their coupling functions within a physics-based modeling framework, we present results from two case studies of I-T response: (1) the extreme Halloween events of Oct/Nov 2003, and (2) the St. Patrick's Day event of 2013. In the context of this event we will highlight modeling challenges; implementation within the Space Weather Technology, Research and Education Center’s (SWx TREC) accelerated R2O-O2R testbed; and expectations for improving I-T forecasting capabilities. 14:45 The Spire TEC Environment Assimilation Model (STEAM) Angling, M et al. Oral M. J. Angling[1], F-X Bocquet[1], G. Olivares-Pulido(1), K. Nordstrom[1], Vu Nguyen[2], T. Duly[2], V. Irisov[2], O. Nogues-Correig[2], L. Tan[3], T. Yuasa[3], D. Masters[2] (1) Spire Global UK Ltd., Glasgow, UK; (2) Spire Global, Inc., Boulder, USA; (3) Spire Global Singapore PKE Ltd., Singapore The ionosphere can affect a wide range of radio frequency (RF) systems operating below 2 GHz. One option for mitigating these effects is to produce assimilative models of the ionospheric density from which products can be derived for specific systems. Such models aim to optimally combine a background model of the ionospheric state with measurements of the ionosphere. This approach is analogous to the use of numerical weather prediction in the meteorological community, and has been evolving for ionospheric use for the last 10 to 15 years. Published research has demonstrated the utility of this approach. However, obstacles to providing effective data products remain due to the sparseness of ionospheric data over large parts of the world and the timeliness with which data are available. Spire is working to overcome these issues through the use of its large constellation of satellites that can measure Total Electron Content (TEC) data in both zenith looking and radio occultation (RO) geometries and its large ground station network that will allow low data latency. Spire data will be combined with an innovative data assimilation model (the Spire TEC Environment Assimilation Model, STEAM) to provide accurate and actionable ionospheric products. Data assimilation is required to overcome the limitations and assumptions of the traditional Abel Transform analysis of RO data (i.e., spherical symmetry; transmitter and receiver in free space and the same plane) and to effectively combine RO data, topside data, ground-based GNSS data, and other sources of ionospheric information (i.e., ionosondes). STEAM uses a 4D Local ensemble transform Kalman Filter (LETKF). As with other ensemble methods, the LETKF uses an ensemble of models to approximate the background error covariance matrix. However, the LETKF provides a more efficient way to solve the ensemble equations. Furthermore, 4D operation permits the use of data with varying latency. Localisation means that grid points are only modified by data within a local volume; this restricts spurious long-range spatial correlations and means that the ensemble only has to span the space locally. The LETKF transforms the problem into ensemble space which makes each grid point independent, resulting in an algorithm that is easily parallelised. This paper will describe the data collection and processing chain, the data assimilation model, and plans for the ongoing development of the combined system. 15:00 Data Assimilation techniques for space weather: how to deal with a very sparse and limited coverage of the solar system Millas, D et al. Oral D. Millas[1], B. Laperre[2], M.E. Innocenti[3], J. Raeder[4], G. Lapenta[5] [1, 2, 3, 5] CmPA, Department of Mathematics, KU Leuven, [3] nterstellar and Heliospheric Physics, Jet Propulsion Laboratory, [4] Institute for the Study of Earth, Oceans, and Space, University of New Hampshire Data assimilation (DA) has been a great asset for meteorological forecasting. It can have a similar beneficial impact on space weather but there are two critical differences. First, contrary to meteorological applications on weather on Earth, Space weather is a hard driven problem, where the solar driver imposes its supersonic and superalfvenic wind on the rest of the system. The propagation of perturbations in a global circulation model of the Earth is very different form that of the solar wind. We need to assess this difference, understand its implications and use them to our advantage. A hard driven problem might be easier than the chaotic global Earth model. Second, there are weather stations all over the globe and several meteorological satellites monitoring the whole planet. This wealth of data compares with images at 4k resolution of the sun and one point of measurement: L1. This severely limits the ability to modify a solar wind model and its disturbances, CMEs, flares and SEPs. Again we need to assess what is possible with the very limited resources and make the best use of them. And also use of knowledge of the model sensitivity to design future realistic missions that might give us the best opportunity to improve the forecasting power of our models. In this study, we combine DA methods with a series of MHD simulations of two typical space weather forecast scenarios. First, the propagation of a CME against the background solar wind, in different scenarios. The CMEs are randomly parametrized (within an interval of typically observed values) to create an ensemble of virtual data; we then examine their effect on different points between the Sun and the Earth, in different times, using representers analysis'' technique . Second, the response of the Earth magnetosphere to varying solar wind drives at L1. In this case we consider the sensitivity of the response in different locations and times. For the simulations, we used PLUTO, EUHFORIA and OpenGGCM. We discuss possible improvements via direct observations and present potential applications of this procedure in the planning of future space missions. \textit{This work has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 776262 (AIDA, www.aida-space.eu)}