1
Contemporary Auroral Tomography Techniques Michael Hirsch, Joshua Semeter Boston University Center for Space Physics Abstract We discuss our approach to high spatial and temporal res- olution auroral tomography, along with an exploration of the observation tradespace in time and space of ground- based observations. We discuss the modeling and process- ing effort relevant to answering science questions on small and large time scales via 100+ frame/second optical auro- ral observations. Tightly time synchronized observations from two or more sensitive cameras enables tomographic reconstruction using a first principles physical model yield- ing new insight into the fine dynamics of primary electron precipitation into the ionosphere down to the ten millisec- ond scale. The High Speed Tomography (HiST) rapidly redeployable instrument contributes to the global synoptic perspective by providing multi-year persistent observations with little user intervention needed due to our OpenCV- based algorithms. Transformative helioscience system ob- servations over solar cycle scales requires systems that take a different approach to systems engineering than legacy sys- tems that assumed frequent human interaction or main- tenance. The techniques we use to study the fine scale spatio-temporal dynamics of the magnetospheric drivers of the aurora can be adapted to other instruments and meta- instruments studying magnetosphere-ionosphere coupling. Introduction The HiST system is part of the leading edge of ionospheric data collection and analysis. Joint sensing of the ionosphere by use of diverse coordinated distributed sensing create meta- instruments where far more can be learned about the iono- sphere than by simply taking the “sum” of isolated measure- ments. The rapidly relocatable HiST system contributes to aeronomy observations of the finest ground-observable auroral features. Auroral tomography is carried out by two or more cameras pointed at a common auroral altitude region, typically set to overlap as much as geometrically possible in the altitude range of z = 90..300 km. Figure 1: Viewing geometry for three cameras at x ∈{0, 3, 10} km. We now turn to a discussion of our preliminary error analysis and the methods used to filter through terabytes of auroral video in a remote autonomous fashion, saving a great deal of financial and human resources. Simulation Error The amount of error in the data inversion is currently estimated by using a Gaussian fit to the data in the neighborhood of the precipitation intensity peak. The peak is detected as the maximum of the precipitation intensity. Figure 2: Simulation error as measured by 2-D Gaussian Levenberg-Marquardt fitter, for E 0 =1keV. Methods Inversion The inversion algorithm has been described in [Hirsch 2015]. Assuming the observation process can be described and dis- cretized with the Fredholm Integral of the First Kind, we use the viewing geometry of Fig. 1 in the projection matrix L. The particle penetration model generating filtered excitation rates is encapsulated in kernel T. The unknown precipitation intensity at the top of the ionosphere that we seek is Φ top . The ground-observable brightness B is inverted to obtain estimated precipitation intensity ˆ Φ top . LTΦ top = B (1) We estimate particular characteristics of ˆ Φ top based on ob- served brightness B using the L-BFGS-B algorithm with cri- teria ˆ Φ top (B ⊥ ,E ) = argmin Φ ||B - LT ˆ Φ top || 2 = argmin Φ ||B - ˆ B|| 2 (2) ˆ Φ top (B ⊥ ,E ) is the location of peak electron precipitation inten- sity in the direction perpendicular to the geomagnetic field and the characteristic energy. Our simulated camera has an expo- sure of 10 ms corresponding to a 100 Hz frame rate. The simu- lation forward model runs at a 2 ms time step, based on known characteristics of dispersive Alfvenic aurora as confirmed by previous ground observations. The result of the 2 ms forward model with a 10 ms exposure is a smeared brightness observed, from which we use a 2-D fitter to estimate ˆ Φ top (B ⊥ ,E ) on 10 ms scale. Machine Vision Algorithm This algorithm (manuscript in preparation) allows filtering many terabytes per day to detect the highly time dynamic auroral events of interest. Image Set Optical Flow Estimation Binary Threshold a< ||u, v || 2 <b Median Filter Erosion Closing Blob analysis Area threshold Record detections Figure 3: Auroral detection algorithm block diagram. Key Result Our inversions of ground-observed auroral optical intensity estimate the characteristic energy E 0 and the location in the directional perpendicular to the geomagnetic field B ⊥ with error on the order of 15%. Our observations are carried out with a network of two or more tightly GPSDO synchronized Electron Multiplying CCD (EMCCD) cameras. Conclusion and Future Work These preliminary results show that tightly time-synchronized cameras can provide access to primary electron precipitation characteristics with spatial and temporal resolution not oth- erwise readily available. As shown in [Semeter 2012], a high- resolution auroral network must be sufficiently close spaced to allow resolving fine B ⊥ structure, and as shown in [Hirsch 2015], a B ∥ physics model such as TRANSCAR can be used to regularize the inherently poorly-observed direction parallel to the geomagnetic field. Our current results and constraints in- herent to the Poker Flat Research Range indicate that sites at 3 km and 10 km spacing give serviceable estimates of the peak in precipitation intensity with regard to the B ⊥,0 location and E 0 characteristic energy. The system is in preparation for re- deployment in Fall 2015 in ruggedized, self-contained outdoor cabinets with integral environmental control. We have work in progress on enhancing the data inversion by incorporation of data from other optical and radar sensors. References [1] P.-L. Blelly et al. “8-moment fluid models of the terrestrial high latitude ionosphere between 100 and 3000 km”. In: Hand- book of the Aeronomical Models of the Ionosphere. Ed. by B. Schunk. CASS, Utah State University, USA: Solar-Terrestrial Environ- ment Program (STEP), 1996. http://scostep.apps01.yorku.ca/ wp-content/uploads/2010/10/ionospheric-models.pdf. [2] H. Dahlgren et al. “The optical manifestation of dispersive field-aligned bursts in auroral breakup arcs”. In: J. of Geophys. Res. (2013). [3] Morales, J. and Nocedal, J. “Remark on Algorithm 778: L-BFGS-B: Fortran Subroutines for Large-scale Bound Constrained Optimization”. In: ACM Trans. Math. Softw. 38.1 (Dec. 2011), 7:17:4. [4] J. Semeter. “Coherence in Auroral Fine Structure”. In: AGU, 2012. [5] Hirsch, M. et al. “Reconstruction of Fine Scale Auroral Dynamics”. In: IEEE Trans. Geo. and Rem. Sens. [In Review]. 2015. [6] Y.-M. Tanaka et al. “Feasibility study on Generalized-Aurora Computed Tomography”. In: Annales Geophysicae 29.3 (2011), pp. 551562. [7] M. Zettergren et al. “Optical estimation of auroral ion upflow: 2. A case study”. In: Journal of Geophysical Research: Space Physics 113.A7 (2008). [8] M. Zettergren et al. “Optical estimation of auroral ion upflow: Theory”. In: Journal of Geophysical Research: Space Physics 112.A12 (2007) Acknowledgements This work was funded by the National Science Foundation Atmosphere and Geospace Science Directorate under Grant 1216530 and Grant 1237376. Thanks to H. Murato and G. Thayer with the Boston University Scientific Instrument Facility for mechanical design and build aspects of this system. CEDAR June 2015 Poster # ITIT-08
