Real-Time UAF Eulerian Parallel Polar Ionosphere Model


Introduction

The ionospheric forecast provided on this page contributes to the National Space Weather Program. Initiated as a part of the first UARC Joint Global Ionospheric Campaign in April, 1997, this WWW-page is active since then. The project is targeted for developing the real-time modeling capabilities using the period-specific geophysical inputs for the space weather forecasts and nowcasts. The critical for the run inputs are available in real time from the WWW-sites of the Space Environment Center of NOAA.

This high-resolution real-time ionospheric forecasting project is based on a first principles polar ionosphere model. The ionospheric maps are prepared with the Eulerian Parallel Polar Ionosphere Model (UAF EPPIM) (1) developed at the University of Alaska Fairbanks and the Arctic Region Supercomputing Center (ARSC). The model inputs, major governing modules and equations, and, finally, outputs, are shown on the folowing flow chart (dotted arrows represent physically existing mechanisms not represented in the model).

The UAF Eulerian Parallel Polar Ionosphere Model code exists in two versions. A scalar version is oriented for a single processor workstation. The parallel version employes capabilites of the high-performance multi-processor computational platforms. For instance, the parallel model version runs with resolutions up to 10x10x10 km on the ARSC parallel supercomputers . Snapshots of distribution of electron density at 290 km of altitude obtained in a series of model runs with identical geophysical conditions and different horizontal resolutions are shown on the following plots (winter solstice, 5.00 UT, solar minimum with F10.7 =100, moderate geomagnetic activity at Kp=2o, weakly positive Bz).

Magnifications of the tongue-of-ionization zone include isocontours drawn at every 0.1 of Log(Ne), which corresponds to a factor of x1.26 for density difference for each neighboring contours. The frames are plotted in the geographic latitude-longtitude frame. On the first plot the frames are limited to the boundary of 50N of the geomagnetic latitude, which in the geographic frame has an oval-like shape. The plots demonstrate how the better computational resolution results in increased solution fidelity, even though the model uses the same limited resolution inputs in each run. This conclusion emphasizes an importance of the horizontal model resolution for the real-time forecasts and explains the authors' attention to the resolution parameter in the following discussion.

As well as the parallel code version, the scalar model code for single processor architectures is computationally highly optimized. As a result, it is capable of running in real-time on a dedicated platform. Depending on the available computational resources, the horizontal resolution is changeable for the real-time runs in 110x110 km to 20x20 km range, supporting the time resolution of 5 minutes. The current real-time run is performed at 30x30 km of horizontal resolution at the dedicated SGI Octane workstation equipped with 300 MHz R12K CPU. The workstation for this project is provoded by the the Arctic Region Supercomputing Center.

A flow-chart of the UAF ionospheric model real-time run shows basic inputs/outputs and the real-time synchronization scheme. The model run is performed in a continuos manner. The periods of active run (230-270 sec) are complemented with some idling time to keep the synchronization interval of 300 sec (5 minutes), coinsiding with the model time step. The model horizontal resolution is adjusted to the highest possible value supported by the available computational resources. Several FTP-processes for fetching the updated inputs exist as parallel UNIX tasks. Thus, the waiting time for FTP of the order of tens of seconds is not intefering with the computations. Altogehter, this approach keeps the hosting platform CPU time usage at a level up to 90%, which essentially requires a dedicated mode.

The UAF model uses as inputs the current indeces of solar and geomagnetic activity available in real time at the Space Environment Center WWW-sites. The inputs are refreshed by the automatic FTP-processes, respectively, once a day (solar activity) and once an hour (US AF/NOAA sinoptic evaluation of the current geomagnetic Kp index). The real-time Interplanetary Magnetic Field (IMF) data at earlier stages of this project was obtined from WIND satellite and now it is taken from the ACE satellite real-time depository. Averaged for 5-minutes interval, IMF data are obtained for each model time step to determine the ionospheric drift pattern in accordance with the statistical electric field model by Weimer [1995,1996]. This statistical model is continuously responsive to the IMF values and, in the same time, to the seasonally variable tilt angle of the Earth's magnetic dipole.

Due to the upstream position of ACE satellite in the solar wind, the ACE IMF data is a good predictor of the geophysical conditions around the Earth. Since it takes certain time for the solar wind to propagate to the Earth's vicinity from the point of measurements, the information about incoming variations is available in some advance. This advance time varies depending on the current ACE position and the solar wind velocity. Assuming ACE at L1 libration point at ~230 Re, the advance time ranges from about 30-40 minutes for a hight solar wind velocity of 700 km/sec to one and a half hour for a low solar wind velocity of 300 km/sec. Thus, dynamically adjusting the model time for this delay, the real-time run generates a forecast of the ionospheric parameters on average one hour in advance:

Model time = Current time + SW Propagation Delay,

where the non-negative

SW Propagation Delay = Distance to ACE / Solar Wind velocity = Forecast Advance

It follows from the relations above that the model time is always ahead of the current time. This mismatch due to the solar wind propagation delay is not constant, depending mainly upon the solar wind velocity variations. The distance to ACE varies much slower and can be assumed constant for weeks- long periods of time.

The idling period in the synchronization scheme is used to dynamically adjust the delay time --or, from forecasting standpoint, advance time-- to a variable solar wind speed. The delay time is inversely proportional to the solar wind velocity. To match the delay time variations, the idling time can be temporaly reduced to zero to increase the delay to the desired level. This "fast" run mode is invoked to accomodate larger delays for the slow solar wind. By contrast, the idling time can be significantly increased for decreasing a shift in the model time with respect to the current time. This "slow" run mode matches an increase of the solar wind velocity.

The other parameters of the run are as follows. The UAF EPPIM vertical step is 10 km with 80-500 km range of the altitude coverage. The model uses Cartesian frame and metrics of the Azimuthal Equidistant Projection to compensate for distortion of the Earth spherical geometry. The model covers the entire polar region, an area of 9100x11000 km, selected to accomodate into the Cartesian geometry a region northward of 50N of both geomagnetic (dashed circle on the coverage plot) and the geographic latitudes (solid circle). During the run, a solution of the governing equations is obtained for the current time step on the entire Eulerian co-rotating mesh, visualized, and transfered to the Web-server in a variety of formats. Altogether, this approach allows for effective use of the UAF Eulerian Parallel Polar Ionosphere Model as a space weather forecasting tool.


Back to the Ionospheric Model main page.