Space weather forecasting with a Multimodel Ensemble Prediction System (MEPS)

Transcription

1 PUBLICATIONS RESEARCH ARTICLE Special Section: Ionospheric Effects Symposium 2015 Key Points: We created a Multimodel Ensemble Prediction System (MEPS) for Earth space based on different models The MEPS ensemble is composed of seven physics-based data assimilation models The goal of the MEPS program is to improve space weather forecasting with ensemble modeling Correspondence to: R. W. Schunk, robert.schunk@usu.edu Citation: Schunk, R. W., et al. (2016), Space weather forecasting with a Multimodel Ensemble Prediction System (MEPS), Radio Sci., 51, , doi:. Received 30 NOV 2015 Accepted 24 JUN 2016 Accepted article online 30 JUN 2016 Published online 27 JUL American Geophysical Union. All Rights Reserved. Space weather forecasting with a Multimodel Ensemble Prediction System (MEPS) R. W. Schunk 1, L. Scherliess 1, V. Eccles 1, L. C. Gardner 1, J. J. Sojka 1, L. Zhu 1,X.Pi 2, A. J. Mannucci 2, M. Butala 2, B. D. Wilson 2, A. Komjathy 2, C. Wang 3, and G. Rosen 3 1 Center for Atmospheric and Space Sciences, Utah State University, Logan, Utah, USA, 2 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA, 3 Department of Mathematics, University of Southern California, Los Angeles, California, USA Abstract The goal of the Multimodel Ensemble Prediction System (MEPS) program is to improve space weather specification and forecasting with ensemble modeling. Space weather can have detrimental effects on a variety of civilian and military systems and operations, and many of the applications pertain to the ionosphere and upper atmosphere. Space weather can affect over-the-horizon radars, HF communications, surveying and navigation systems, surveillance, spacecraft charging, power grids, pipelines, and the Federal Aviation Administration (FAA s) Wide Area Augmentation System (WAAS). Because of its importance, numerous space weather forecasting approaches are being pursued, including those involving empirical, physics-based, and data assimilation models. Clearly, if there are sufficient data, the data assimilation modeling approach is expected to be the most reliable, but different data assimilation models can produce different results. Therefore, like the meteorology community, we created a Multimodel Ensemble Prediction System (MEPS) for the Ionosphere-Thermosphere-Electrodynamics (ITE) system that is based on different data assimilation models. The MEPS ensemble is composed of seven physics-based data assimilation models for the ionosphere, ionosphere-plasmasphere, thermosphere, high-latitude ionosphere-electrodynamics, and middle to low latitude ionosphere-electrodynamics. Hence, multiple data assimilation models can be used to describe each region. A selected storm event that was reconstructed with four different data assimilation models covering the middle and low latitude ionosphere is presented and discussed. In addition, the effect of different data types on the reconstructions is shown. 1. Introduction The seven MEPS models in our data assimilation suite are listed in Table 1, but in our initial studies the focus was on the four GAIM models. GAIM stands for Global Assimilation of Ionospheric Measurements for USU models (Gauss-Markov (GM) and full physics (FP)) and Global Assimilative Ionospheric Model for Jet Propulsion Laboratory (JPL)/University of Southern California (USC) models (band limited (BL) and 4DVAR). The GAIM models were selected for this study because they are mature, cover the same middle to low latitude region, have been used independently in several studies, and are able to assimilate the same data types. In comparing ionospheric reconstructions, it was anticipated that different data assimilation (DA) models used to describe the same geophysical event could yield different results, because the different DA models are based on different background physics-based models, data sources, assimilation techniques, and spatial and temporal resolutions. However, the GAIM models may also reveal similar disturbance patterns since all data assimilation techniques attempt to bring models in line with observations. Therefore, we conducted a systematic comparison of the models in order to determine the differences and similarities in the reconstructions. First, the four GAIM models were run the way they normally are run, with different data amounts and types. As expected, there were similarities and differences in the four GAIM reconstructions but that first test established the state of the art [see Schunk et al., 2014a, 2014b]. In this study, we ran the four GAIM models for the March 2013 period with the same data, but we also systematically added new data types and reran the GAIM models to see how the different data types affected the GAIM results. First, the models were run with slant total electron content (TEC) data from 530 ground receivers, then with the slant TEC data and Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) occultation data, and finally with slant TEC data, occultation data, and bottomside profiles from 80 ionosondes. The data used in this study were acquired from the web: ground-based Global SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1157

2 Table 1. Data Assimilation Models Model GAIM-Band Limited (BL) GAIM-Gauss-Markov (GM) GAIM-4DVAR GAIM-full physics (FP) Middle to low electro-da IDED-DA GTM-DA Region Middle to Low Latitude Ionosphere Middle to Low Latitude Ionosphere Middle to Low Latitude Ionosphere with Drivers Middle to Low Latitude Ionosphere-Plasmasphere with Drivers Middle to Low Latitude Ionosphere with Drivers High-Latitude Ionosphere with Drivers Global Thermosphere Model-Data Assimilation Positioning Satellite (GPS) slant total electron content (TEC) measurements (ftp://cddis.gsfc.nasa.gov/pub/ gps/data/daily, ftp://ftp.ngs.noaa.gov/cors/rinex, ftp://garner.ucsd.edu/pub/rinex, and ftp://data-out.unavco. org/pub/rinex/obs); ionosonde/digisonde Standard Archiving Output (SAO) data files (ftp://ftp.ngdc.noaa. gov/ionosonde/data/); Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) radio occultation (RO) absolute total electron content from the Precision Omnidirectional Dipole (POD) antennas ( and Defense Meteorological Satellite Program (DMSP). 2. GAIM Models The GAIM-GM model [Schunk et al., 2004a, 2004b; 2005a, 2005b; Scherliess et al., 2004, 2006, 2011] uses a physics-based global ionosphere model and a Kalman filter as a basis for assimilating a diverse set of measurements. The ionosphere model is the Ionosphere Forecast Model (IFM), which covers the E region, F region, and topside ionosphere up to 1500 km and takes account of five ion species (NO +,O + 2,N + 2,O +, and H + ). The IFM is based on a numerical solution of the plasma continuity, momentum, and energy equations [Schunk, 1988; Sojka, 1989; Schunk et al., 1997], and it takes account of a myriad of chemical, physical, and transport processes, and the displacement between the magnetic and geographic poles. The IFM calculates 3-D, time-dependent density, drift velocity, and temperature distributions for the plasma species. In GAIM-GM, the IFM N e distribution constitutes the background field on which perturbations are superimposed based on the data and their errors. The perturbations and associated errors evolve over time with a Gauss- Markov process [see Scherliess et al., 2006]. The GAIM-FP model uses an ensemble Kalman filter technique and rigorously evolves the ionosphere and plasmasphere electron density field and its associated errors using the full physical model [Scherliess et al., 2004, 2009, 2011, 2015; Schunk et al., 2004b, 2005b]. GAIM-FP is based on the Ionosphere-Plasmasphere Model (IPM), which is constructed from a numerical solution of the ion-electron transport equations. The IPM calculates 3-D time dependent, density, drift velocity, and temperature distributions along B for six ions (NO +,O + 2,N + 2,O +,He +, and H + ) in the ionosphere and plasmasphere for equatorial crossing altitudes from 90 to 30,000 km. The plasma is allowed to drift across B in response to corotational, dynamo, and storm time electric fields [see Schunk et al., 2003]. The different data sources are assimilated in GAIM-FP via an ensemble Kalman filter technique and the model provides 3-D plasma distributions versus time. It also provides selfconsistent distributions for the global drivers, including neutral winds and densities, and equatorial electric fields [see Scherliess et al., 2009]. The GAIM-GM and GAIM-FP models have been used in numerous validation and sensitivity studies in order to determine their performance characteristics and accuracy for both global and regional reconstructions [Scherliess et al., 2006; Thompson et al., 2006, 2009; Sojka et al., 2007; Jee et al., 2007, 2008; Zhu et al., 2006, 2012; Schunk et al., 2011; Gardner et al., 2014a, 2014b]. The GAIM-JPL/USC models were developed jointly at the University of Southern California and Jet Propulsion Laboratory [Hajj et al., 2000; Rosen et al., 2001; Pi et al., 2003, Wang et al., 2004; Hajj et al., 2004]. They incorporate a physics-based 4-D global ionospheric model, an observation operator, and data assimilation modules, including Band-Limited Kalman filter and 4DVAR approach. GAIM-JPL/USC physics models utilize an Eulerian approach and numerically solve the plasma hydrodynamic equations including ionosphere and plasmasphere on a fixed 3-D geomagnetic grid (IGRF-like). With state-of-the-art empirical models of solar EUV radiation, thermosphere, winds, electric fields, and auroral precipitation, the GAIM models compute the densities of H +,He +,O +,N + 2,NO +,O + 2, and electrons. To model ionospheric weather, the GAIM models SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1158

3 Figure 1. Snapshots of global N m F 2 distributions from the GAIM models at 21 UT on the storm day (17 March 2013). (top row) The results for the GAIM models with self-consistent drivers (left plot is 4DVAR, right plot is FP) and (bottom row) the models that do not calculate the drivers (left plot is BL, middle plot is GM). The right plot shows the background IPM model used in GAIM-FP. assimilate observations, particularly GPS data collected from ground- and space-based receivers and ionosonde data. The data assimilation attempts to minimize model deviations from real-world conditions by covariance estimation of ionospheric state or model driver estimation. In GAIM-4DVAR, an adjoint method and parameterizations for model drivers have been implemented [Rosen et al., 2001; Pi et al., 2003; Wang et al., 2004]. With the adjoint method, the computational burden remains essentially fixed and independent of the number of parameters. In addition, the driver parameterizations have been applied to 4DVAR driver estimation that help to reduce the number of parameters to be estimated. For example, in 4DVAR, the E B vertical drift velocity at the magnetic equator is formulated as an empirical drift plus a deviation. The deviation to be estimated is a function of local solar time using a group of periodic cubic spline basis functions. In addition, the thermospheric wind is also parameterized on a separate 3-D grid constructed with relative coarser resolutions for wind estimation. This parameterization ensures that the adjustment of the wind in each model voxel is only affected by local wind parameters. Similar parameterization is performed for ion production as well. The 4DVAR approach has been used in several data assimilations to estimate plasma E B drift, wind, and EUV flux at low and middle latitudes [e.g., Pi et al., 2003, 2004, 2008]. Validation of GAIM-JPL/USC is routinely conducted using various independent data sources, including TEC, density, h m F 2, and plasma drift measurements obtained from TOPEX and Jason altimeters, incoherent scatter radar, ionosonde, radio occultation inversion, and GPS data that are deprived from assimilations [e.g., Mandrake et al., 2005; Pi et al., 2009a,2009b; Komjathy et al., 2010]. 3. GAIM Ionosphere Reconstructions The reconstructions shown in what follows are for the March 2013 period; the first day was quiet (Kp ~ 2) but a persistent storm occurred on the second day (Kp ~ 6). For our first test case, all four GAIM models assimilated only slant TEC from 530 ground GPS receivers. Figure 1 shows snapshots of the resulting global N m F 2 distributions from the GAIM models at 21 UT on the storm day (17 March 2013). Figure 1 (left column) shows the JPL/USC results and Figure 1 (middle column) shows the USU results. Figure 1 (top row) shows the results for the GAIM models with self-consistent drivers and Figure 1 (bottom row) is for the models that do not calculate the drivers. The third plot on the right in Figure 1 (bottom row) shows the result from the background IPM model used in GAIM-FP. A comparison of the results indicates that there are differences in the magnitude of the equatorial anomaly and some differences in the longitudinal extent and width of the anomaly. However, the four models show a similar enhanced N m F 2 feature in the Southern Hemisphere at latitudes poleward of 30 south. Figure 2 shows the comparison of the corresponding h m F 2 distributions, and as with the density distributions, there are differences that can be as large as km near the magnetic equator. SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1159

4 Figure 2. Same format and time as for Figure 1 but for h m F 2 distributions. It is apparent that the differences in the GAIM reconstructions can be significant even in the regions where there is a lot of data. Part of the problem is that only ground GPS/TEC data were assimilated in the case shown in Figures 1 and 2, and this leads to uncertainties in N m F 2 and h m F 2 (see discussion that follows). In addition, the four GAIM models were just run once with no fine tuning to correct for uncertainties. This is the procedure that is needed for ensemble modeling with different data assimilation models if specifications and forecasts are desired. However, to provide insight as to why there are differences in the GAIM reconstructions, in what follows we provide further details about the two GAIM models that do not deduce the ionosphere drivers, the two GAIM models that do determine the drivers, and the uncertainties associated with driver determination. For the two data assimilation models that do not deduce the self-consistent drivers (GAIM-GM and GAIM- BL), the time-dependent, three-dimensional, electron density distributions calculated by the background ionosphere models are adjusted to be consistent with the various measurements to within the uncertainties of the background ionosphere model and measurements. An exact fit to the individual measurements is not achieved, but rather a simultaneous overall agreement is achieved. The influence of a specific measurement is propagated away from its location via a statistical process for GAIM-GM [Scherliess et al., 2006] and via a band-limited approach for GAIM-BL [Hajj et al., 2000; Rosen et al., 2001; Pi et al., 2004]. The two data assimilation approaches are not expected to produce identical results because the background ionosphere models, assimilation techniques, and error estimates are different in GAIM-GM and GAIM-BL. For the two data assimilation models that calculate the self-consistent drivers (GAIM-FP and GAIM-4DVAR), the time-dependent, three-dimensional, electron density distributions and the self-consistent ionosphere drivers (neutral winds, neutral density profiles, production rates, and equatorial electric fields) are adjusted self-consistently to obtain agreement with the various measurements, to within the uncertainties of the background ionosphere model, deduced drivers, and measurements. As with the GAIM-GM and GAIM-BL models, an exact fit to the individual measurements is not achieved; a simultaneous overall agreement is achieved. For GAIM-FP and GAIM-4DVAR, the influence of a specific measurement is propagated away from its location via the physics in the background ionosphere models [see Scherliess et al., 2009; Pi et al., 2003]. The drivers are deduced from ionospheric observations and this can lead to new reconstruction errors if the physics in the background ionosphere models is not complete [Schunk et al., 2011]. For example, if the background ionosphere model does not include bubble physics, assimilating electron density measurements that pass through plasma bubbles can lead to significant errors with regard to the deduced drivers and reconstructed electron density distribution. Also, the determination of the drivers is not unique. The peak electron density (N m F 2 ), peak height (h m F 2 ), and bottomside profile are affected by neutral winds, equatorial electric fields, and neutral composition changes. At the magnetic equator, the electric field dominates the h m F 2 variation and this helps determine the electric field. An asymmetry in the equatorial ionization anomaly peaks helps determine the neutral wind in the magnetic meridian. But the neutral composition causes uncertainties. However, a comparison of the deduced drivers would provide insight into what causes some of the SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1160

5 Figure 3a. Snapshots of GAIM-GM reconstructions for N m F 2 (left column) and h m F 2 (right column) for 21 UT on the storm day (March 17, 2013). For the top row, only slant TEC from 530 ground GPS receivers were assimilated, for the middle row both slant TEC from 530 ground GPS receivers and COSMIC occultation data were assimilated, and for the bottom row bottom-side electron density profiles from 80 globally distributed digisondes were assimilated in addition to the two other data types. Figure 3b. The N m F 2 (left column) and h m F 2 (right column) distributions from GAIM-GM when the GPS/TEC data assimilation results (Figure 3a, top row) are subtracted from the Figure 3a middle and bottom rows. Absolute differences are shown to highlight the regions where the model results are significantly different, since with a percentage plot, relative differences would be highlighted. SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1161

6 Figure 4a. Snapshots of GAIM-FP reconstructions for N m F 2 (left column) and h m F 2 (right column) for 21 UT on the storm day (March 17, 2013). For the top row, only slant TEC from 530 ground GPS receivers were assimilated, for the middle row both slant TEC from 530 ground GPS receivers and COSMIC occultation data were assimilated, and for the bottom row bottom-side electron density profiles from 80 globally distributed digisondes were assimilated in addition to the two other data types. Figure 4b. The N m F 2 (left column) and h m F 2 (right column) distributions from GAIM-FP when the GPS/TEC data assimilation results (Figure 4a, top row) are subtracted from the Figure 4a middle and bottom rows. Absolute differences are shown to highlight the regions where the model results are significantly different, since with a percentage plot, relative differences would be highlighted. SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1162

7 differences between the models. There are also uncertainties in the measurements. For example, ground GPS/TEC measurements typically need to be corrected for receiver and satellite biases and account has to be taken of how much of the TEC is in the topside ionosphere and plasmasphere. The second test case involves the GAIM-GM and GAIM-FP models, and it shows how the corresponding GAIM reconstructions are modified as additional data types are included in the data assimilation scheme. Figure 3a shows snapshots of GAIM-GM reconstructions for N m F 2 (left column) and h m F 2 (right column) for 21 UT on the storm day (17 March 2013). For Figure 3a (top row), only slant TEC from 530 ground GPS receivers were assimilated, for Figure 3a (middle row) both slant TEC from 530 ground GPS receivers and COSMIC occultation data were assimilated, and for Figure 3a (bottom row) bottomside electron density profiles from 80 globally distributed digisondes were assimilated in addition to the two other data types. To better display the effect of the different data types on the reconstructions, Figure 3b shows the N m F 2 (left column) and h m F 2 (right column) distributions when the GPS/TEC distributions (Figure 3a, top row) are subtracted from Figure 3a (middle and bottom rows). The COSMIC occultation data have a significant effect over the oceans, but it is also important over land (Figure 3b, top row). In Figure 3b (bottom row), the effect of the digisonde data is included with the effect of the occultation data. However, it is still evident that the digisonde data are important and its influence is clearly visible in the h m F 2 distribution (Figure 3b, bottom row, right plot). Note that the effect of the digisonde data is transported to regions away from the digisonde sites, and even over the oceans, by the data assimilation model. For example, h m F 2 at the equator is significantly modified in the pacific sector by the assimilation of digisonde data in the west continent and its interplay with the other data types. Figures 4a and 4b show snapshots of the GAIM-FP parameters for 21 UT on the storm day (17 March 2013). The format and color scales are the same as for the GAIM-GM results shown in Figures 3a and 3b. A comparison of the GAIM-GM and GAIM-FP results indicates that there are noticeable differences in the ionospheric reconstructions and that the different data sources affect the GM and FP reconstructions differently. For example, the anomaly peaks are more sharply defined and have a greater separation from the equatorial minimum in the GAIM-FP assimilation than those in the GAIM-GM assimilation. With regard to the relative merits of the three data types, it is evident that the more data that are assimilated into a physics-based model, the better the reconstruction is expected to be. However, it is also important to assimilate different data types because they affect the reconstruction in different ways. Ground GPS/TEC data are abundant, but they are only available over land. Also, as noted earlier, TEC is an integrated quantity that extends from the ground receiver to the GPS satellites. Assimilated line-of-sight TEC measurements intercept the ionosphere simultaneously from different elevation and azimuth angles, forcing the model to adjust ionospheric densities in a 3-D region to fit the data through data assimilation algorithms. This can result in density profile changes, and at least, N m F 2 changes in the modeled region. The effect of ground TEC measurements on h m F 2 is complicated based on our observations, and it is still a research topic. The COSMIC radio occultation data provide slant TEC between the COSMIC satellites and the GPS satellites. Its main attribute is to provide information over the vast ocean areas. The digisonde data provide bottomside electron density profiles at specific locations. The profiles are especially important in helping to deduce the ionosphere drivers (neutral winds, composition, and equatorial electric fields). 4. Summary As expected, there are important differences when the four GAIM DA models are used to reconstruct the same geophysical event using the same data. However, it should be noted that the four GAIM models were just run once with no fine tuning to correct for uncertainties, which is the way they would be run for specification and forecasting. In an effort to better understand the differences and improve the individual models, the next step is to compare the background ionosphere models (IFM, IPM, and JPL ionosphere), the way data errors are handled, and the strengths/limitations of the data assimilation schemes. A comprehensive validation effort is also needed. However, even after improvements, the four models are not expected to agree. A similar situation occurred in terrestrial weather modeling, where ensemble modeling with different weather models produced better forecasts than any one of the individual models. This information provides the motivation for using the four GAIM models in a Multimodel Ensemble Prediction System (MEPS) for space weather specifications and forecasts. SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1163

8 Acknowledgments The research was supported by the NASA/NSF Space Weather Modeling Collaboration program via NSF grant AGS to Utah State University. The research conducted at the Jet Propulsion Laboratory, California Institute of Technology, is under a contract with the National Aeronautics and Space Administration. The model output from these simulations is available on the USU CASS computer system. (Contact Larry Gardner at larry. gardner@usu.edu for information.) References Gardner, L. C., R. W. Schunk, L. Scherliess, L. Zhu, and J. J. Sojka (2014a), Ionospheric reconstruction for various solar, seasonal, and geomagnetic conditions obtained from the Global Assimilation of Ionospheric Measurements Gauss Markov (GAIM-GM) model, Proceedings of the Institute of Navigation International Technical Meeting, San Diego, Calif. Gardner, L. C., R. W. Schunk, L. Scherliess, J. J. Sojka, and L. Zhu (2014b), Global assimilation of ionospheric measurements (GAIM) Gauss Markov (GM) Model: Improved specifications with multiple data types, Space Weather, 12, , doi: /2014sw Hajj, A. G., L. C. Lee, X. Pi, L. J. Romans, W. S. Schreiner, P. R. Straus, and C. Wang (2000), COSMIC GPS ionospheric sensing and space weather, Terrs. Atmos. Oceanic Sci., 11, 235. Hajj, G. A., B. D. Wilson, C. Wang, X. Pi, and I. G. Rosen (2004), Data assimilation of ground GPS total electron content into a physics-based ionospheric model by use of the Kalman filter, Radio Sci., 39, RS1S05, doi: /2002rs Jee, G., A. G. Burns, W. Wang, S. C. Solomon, R. W. Schunk, L. Scherliess, D. C. Thompson, J. J. Sojka, and L. Zhu (2007), Duration of an ionospheric data assimilation initialization of a coupled thermosphere-ionosphere model, Space Weather, 5, S01004, doi: / 2006SW Jee, G., A. G. Burns, W. Wang, S. C. Solomon, R. W. Schunk, L. Scherliess, D. C. Thompson, J. J. Sojka, and L. Zhu (2008), Driving the TING model with GAIM electron densities: Ionospheric effects on the thermosphere, J. Geophys. Res., 113, A03305, doi: /2007ja Komjathy, A., B. Wilson, X. Pi, V. Akopian, M. Dumett, B. Iijima, O. Verkhoglyadova, and A. J. Mannucci (2010), JPL/USC GAIM: On the impact of using COSMIC and ground-based GPS measurements to estimate ionospheric parameters, J. Geophys. Res., 115, A02307, doi: / 2009JA Mandrake, L., B. Wilson, C. Wang, G. Hajj, A. Mannucci, and X. Pi (2005), A performance evaluation of the operational Jet Propulsion Laboratory/University of Southern California Global Assimilation Ionospheric Model (JPL/USC GAIM), J. Geophys. Res., 110, A12306, doi: /2005ja Pi, X., C. Wang, G. A. Hajj, I. G. Rosen, B. D. Wilson, and G. Bailey (2003), Estimation of E B drift using a global assimilative ionospheric model: An observation system simulation experiment, J. Geophys. Res., 108(A2), , doi: /2001ja Pi, X., C. Wang, G. A. Hajj, G. Rosen, B. D. Wilson, and A. J. Mannucci (2004), Assimilative modeling of low-latitude ionosphere, Proc. IEEE PLANS, Pi, X., V. Akopian, A. J. Mannucci, B. D. Wilson, A. Komjathy, B. A. Iijima, T. F. Runge, and M. A. Dummet (2008), Modeling low-latitude ionosphere using GAIM assimilating GPS data, paper presented at 12 th International Symposium on Equatorial Aeronomy (ISEA-12), Crete, Greece. Pi, X., et al. (2009a) OSSE Using JPL-USC GAIM, COSMIC Workshop, Boulder, Colo., October 28, Pi, X., A. J. Mannucci, B. A. Iijima, B. D. Wilson, A. Komjathy, T. F. Runge, and V. Akopian (2009b), Assimilative modeling of ionospheric disturbances with FORMOSAT-3/COSMIC and ground-based GPS measurements, Terr. Atmos. Ocean. Sci., 20(1), Rosen, I. G., C. Wang, G. Hajj, X. Pi, and B. Wilson (2001), An adjoint method based approach to data assimilation for a distributed parameter model for the ionosphere, paper presented at 40 th Conference on Decision and Control, Inst. of Electr. and Electron. Eng., Orlando, Fla. Scherliess, L., R. W. Schunk, J. J. Sojka, and D. Thompson (2004), Development of a physics-based reduced state Kalman filter for the ionosphere, Radio Sci., 39, RS1S04, doi: /2002rs Scherliess, L., R. W. Schunk, J. J. Sojka, D. C. Thompson, and L. Zhu (2006), Utah State University Global Assimilation of Ionospheric Measurements Gauss-Markov Kalman filter model of the ionosphere: Model description and validation, J. Geophys. Res., 111, A11315, doi: /2006ja Scherliess, L., D. C. Thompson, and R. W. Schunk (2009), Ionospheric dynamics and drivers obtained from a physics-based data assimilation model, Radio Sci., 44, RS0A32, doi: /2008rs Scherliess, L., D. C. Thompson, and R. W. Schunk (2011), Data assimilation models: A new tool for ionospheric science and applications, in The Dynamic Magnetosphere, IAGA Special Sopron Book Series, vol. 3, edited by W. Liu and M. Fujimoto, pp , Springer, Berlin. Scherliess, L., R. W. Schunk, L. C. Gardner, L. Zhu, J. V. Eccles, and J. J. Sojka (2015), The USU-GAIM data assimilation models for ionospheric specifications and forecasts, Proceedings of the 2015 Ionospheric Effects Symposium. Schunk, R. W. (1988), A mathematical model of middle and high-latitude ionosphere, PAGEOPH, l27, Schunk, R. W., J. J. Sojka, and J. V. Eccles (1997), Expanded capabilities for the ionospheric forecast model, Report AFRL-VS-HA-TR , Air Force Research Lab., Space Vehicles Directorate, Hanscom AFB, Mass. Schunk, R. W., J. V. Eccles, J. J. Sojka, D. C. Thompson, and L. Zhu (2003), Assimilation Ionosphere Model (AIM), Final Report, Space Environment Corporation, Providence, UT. Schunk, R. W., L. Scherliess, J. J. Sojka, and D. Thompson (2004a), Global Assimilation of Ionospheric Measurements (GAIM), Radio Sci., 39, RS1S02, doi: /2002rs Schunk, R. W., L. Scherliess, J. J. Sojka, and D. C. Thompson (2004b), USU global ionospheric data assimilation models, Proc. SPIE, 5548, , doi: / Schunk, R. W., L. Scherliess, J. J. Sojka, D. C. Thompson, and L. Zhu (2005a), An operational data assimilation model of the global ionosphere, Proc. Ionospheric Effects Symposium, edited by J. M. Goodman, pp , JMG Assoc., Alexandria, Va. Schunk, R. W., L. Scherliess, J. J. Sojka, D. C. Thompson, and L. Zhu (2005b), Ionospheric weather forecasting on the horizon, Space Weather, 3, S08007, doi: /2004sw Schunk, R. W., L. Scherliess, and D. C. Thompson (2011), Ionosphere data assimilation: Problems associated with missing physics, in Aeronomy of the Earth s Atmosphere and Ionosphere, IAGA Special Sopron Book Series, vol. 2, edited by D. P. M. Ali Abdu and A. Bhattacharyya, pp , Springer, Berlin. Schunk, R. W., et al. (2014a), Multimodel ensemble prediction system for space weather applications, Proceedings of the Institute of Navigation International Technical Meeting, San Diego, Calif. Schunk, R. W., et al. (2014b), Ensemble modeling with data assimilation models: A new strategy for space weather specifications, forecasts and science, Space Weather, 12, , doi: /2014sw Sojka, J. J. (1989), Global scale, physical models of the F region ionosphere, Rev. Geophys., 27, , doi: /rg027i003p Sojka, J. J., D. C. Thompson, L. Scherliess, R. W. Schunk, and T. J. Harris (2007), Assessing models for ionospheric weather specifications over Australia during the 2004 Climate and Weather of the Sun-Earth-System (CAWSES) campaign, J. Geophys. Res., 112, A09306, doi: / 2006JA Thompson, D. C., L. Scherliess, J. J. Sojka, and R. W. Schunk (2006), The Utah State University Gauss-Markov Kalman filter of the ionosphere: The effects of slant TEC and electron density profile data on model fidelity, J. Atmos. Sol. Terr. Phys., 68, Thompson, D. C., L. Scherliess, J. J. Sojka, and R. W. Schunk (2009), Plasmasphere and upper ionosphere contributions and corrections during the assimilation of GPS slant TEC, Radio Sci., 44, RS0A02, doi: /2008rs SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1164

9 Wang, C., G. A. Hajj, X. Pi, I. G. Rosen, and B. D. Wilson (2004), Development of the global assimilative ionospheric model, Radio Sci., 39, RS1S06, /2002RS Zhu, L., R. W. Schunk, G. Jee, L. Scherliess, J. J. Sojka, and D. C. Thompson (2006), Validation study of the Ionosphere Forecast Model using the TOPEX total electron content measurements, Radio Sci., 41, RS5S11, doi: /2005rs Zhu, L., R. W. Schunk, L. Scherliess, and V. Eccles (2012), Importance of data assimilation technique in defining the model drivers for the space weather specification of the high-latitude ionosphere, Radio Sci., 47, RS0L24, doi: /2001rs SCHUNK ET AL. SPACE WEATHER FORECASTING WITH MEPS 1165