Space and Solar Physics (Jay Johnson, organizer)


Jay Johnson (Princeton University Plasma Physics Laboratory)
An Information-Dynamical Approach to Identify Nonlinearity in Magnetospheric Activity

Friday 8:30-9:00, Fountain III

Abstract:

Understanding the dynamical evolution of the Earth's magnetosphere is of practical interest because the magnetosphere occasionally evolves into a disturbed state that can affect the quality of life through large scale damage to power grids, loss of communications, and disruption of satellite-based defense strategy. The magnetospheric dynamics are ultimately driven by the solar wind while various dissipative processes cause the magnetosphere to evolve toward a quiescent state in the absence of strong driving. The magnetospheric dynamics are commonly characterized with various information-dynamical measures to understand dimensionality as well as the most important dependencies among observed plasma and electromagnetic field variables in the coupled solar wind/magnetosphere system. We discuss a method to detect nonlinear dependencies in multivariate time series using mutual information and cumulant-based cost as discriminating statistics. The method is applied to the historical data stream of geomagnetic indices spanning over six solar cycles. We identify nonlinear dependencies using mutual information and cumulant-based cost as discriminating statistics and discuss implications for modeling the magnetosphere and predicting its evolution. We also discuss the relative merits of mutual information and cumulant based cost as discriminating statistics in the context of limited or noisy datasets.


Brian Wilson (Jet Propulsion Laboratory)
Lukas Mandrake (Jet Propulsion Laboratory)
George Hajj (Jet Propulsion Laboratory)
Chunming Wang (University of Southern California)
JPL/USC GAIM: A Real-Time Global Ionospheric Data Assimilation Model

Friday 9:00-9:30, Fountain III

Abstract:

The space weather community has increasingly recognized in recent years the importance of real-time and near real-time (NRT) data streams to support NRT monitoring and forecast of both day-to-day space weather and extreme magnetospheric & ionospheric storm events driven by the Sun. In April 2006 the COSMIC constellation (6 satellites) will launch, enabling an unprecedented continuous, global, 3-dimensional view of the ionosphere using space-based GPS limb scans or occultations. Calibrated limb scans at a data rate of 1 to 10 seconds will be available within 30-120 minutes of real time. In addition, total electron content (TEC) measurements (line-of-sight path delay) are available every second from a worldwide network of 90+ ground-based, dual-frequency GPS sites (streaming GPS network), with additional global coverage every hour from another 100 sites (hourly GPS network). We are in the midst of a revolution in ionospheric remote sensing driven by the illuminating powers of these ground and space-based GPS receivers, new UV remote sensing satellites, and the advent of data assimilation techniques for space weather.

The University of Southern Califronia (USC) and the Jet Propulsion Laboratory (JPL) have jointly developed a Global Assimilative Ionospheric Model (GAIM) to monitor space weather, study storm effects, and provide ionospheric calibration for DoD customers and NASA flight projects. GAIM is a physics-based 3D data assimilation model that uses both variational (4DVAR) and Kalman filter techniques to solve for the ion & electron density state and key drivers such as equatorial electrodynamics, neutral winds, and ion production terms. GAIM can accept as input: TEC measurements from the full worldwide network of 1000+ GPS sites; GPS occultation links from CHAMP, SAC-C, IOX, and the COSMIC constellation; UV limb and nadir scans from the TIMED and DMSP satellites; and in situ data from a variety of satellites (C/NOFS & DMSP). GAIM ingests the multiple data sources, in near real-time if available, updates the 3D ion & electron density grid every 5 minutes, and solves for improved drivers every 1-2 hours. Since our forward physics model and corresponding adjoint model were expressly designed for data assimilation and computational efficiency, all of this can be accomplished on one or two Linux workstations. GAIM density retrievals have been validated by comparisons to vertical TEC measurements from TOPEX & JASON, line-of-sight TEC measurements from independent GPS sites, density versus altitude profiles from ionosondes & incoherent scatter radars, and alternative tomographic retrievals.

The varying latency of the NRT data sources, from 5 minutes to 2 hours, presents a serious challenge to any NRT data assimilation system. On the one hand, one wants to provide a NRT density specification by continuously ingesting the low latency data and performing a 3D density (Kalman filter) update every 5 to 15 minutes. However, the powerful datatypes from the space-based platforms, with latencies up to 2 hours, are necessary to accurately image the ionosphere in 3D and, more importantly, estimate ionospheric drivers that are not currently (directly) measurable, thereby enabling forecast. So on the other hand, one would like to fully utilize all of the data by running the assimilation system 2 hours in the past, and then quickly propagating forward to provide a RT specification and forecast. One potential solution is to run two threads that exchange density state and drivers, but tradeoffs in space & time resolution are inevitable and the computation must be kept tractable. In addition, an operational system must deal with other issues such as varying data latencies, data outages, editing & calibrating data in RT, diagnosing when two datatypes are inconsistent, and rejecting erroneous data using post-fit residual checks. Finally, the accuracy of the nowcast & forecast products must be continuously validated by comparisons to independent measurements, both in NRT and after the fact. Since independent datatypes may not be available in NRT, continuous validation must often be done by withholding part of the input dataset and instead reserving it for validation.

A prototype real-time (RT) GAIM system was first demonstrated in May 2004 and is currently running every day. We will present results from this prototype, describe how we handle the NRT measurement stream, and discuss our approach to the many challenges mentioned above.


Simon Wing (Applied Physics Laboratory, Johns Hopkins University)
Cross-Fertilization between Machine Learning and Space Physics

Friday 9:30-10:00, Fountain III

Abstract:

The underlying physics of many space objects and phenomena is complex and often not well understood, which makes it difficult to develop physics based models or algorithms. However, for many cases, progress can be attained through the use of advanced or even standard machine learning or artificial intelligence algorithms. One of the biggest challenges is to identify and pair the appropriate computing science tools with relevant physical problems. Three representative examples, in which successful cross-fertilizations between space physics and computer science have been achieved, are presented. In the first example, a neural network was developed to classify high frequency (HF) radar returns from the ionospheric irregularities (sometimes called clutters) into those suitable, or not, for further analysis. Such a task is time-consuming and had required human intervention. The dataset assembled for this project has been widely distributed and is known as the "The Johns Hopkins radar dataset" in the machine learning community. Our results suggest that such neural networks could aid many HF radar operations such as frequency search etc. The second example describes a model that is used to classify the source regions of the ions and electrons that rain down upon or precipitate into the upper atmosphere at high-latitudes. One of the manifestations of this precipitation is the aurora. Due to the complexity of the task, the model needs to combine two paradigms: neural networks and an expert system. Finally, the third example describes neural network based models for forecasting global geomagnetic activity indices, Kp and Dst. These indices are useful for a wide range of space applications such as commnunications, navigation systems, satellite health, power grids, commercial air travels, etc.