A. A. Namgaladze, O. V. Martynenko, and A. N. Namgaladze
Polar Geophysical Institute, Murmansk, Russia
The method of numerical simulation of physical processes in the terrestrial upper atmosphere during recent years has taken a prominent position in global geophysical science as an interpretation tool for coordinated programs and observations and, first of all, for incoherent scatter and satellite data. Significant progress in method development and its successive application to solving many problems of interactions within the various regions of the stratosphere-mesosphere-ionosphere-magnetosphere system is mainly due to rapid progress of computer technology, which allows rapid processing of huge amounts of information and solution of more and more complicated problems of mathematical physics by numerical methods.
A global numerical self-consistent model of the terrestrial thermosphere, ionosphere, and protonosphere as a united system was developed by A. A. Namgaladze et al. [1988, 1990, 1991]. Korenkov et al. [1990, 1993] demonstrated the capacity of this model to reproduce adequately observed behavior of the upper atmosphere at least in quiet conditions. Barbolina et al.  and A. N. Namgaladze et al.  developed a modification of the model program to be used on IBM PC AT 386 and IBM PC AT 486 computers. However, the model spatial resolution ( 5o in latitude for the parameters of the ionospheric F2 region and protonosphere and for the electric field potential and 10o for parameters of the thermosphere and lower ionosphere) was obviously insufficient to simulate high-latitude ionospheric disturbances, especially in the auroral and subauroral regions where characteristic latitudinal dimensions of the principal large-scale irregularities (the auroral oval in the ionospheric E region, main ionospheric trough in the F2 region) are about several degrees.
To develop a new modified global model of the terrestrial thermosphere, ionosphere, and protonosphere with an increased spatial resolution which allows using the model as a tool to study high-latitude events in the upper atmosphere is a principal goal of this paper.
The problem of an increase of the spatial resolution of the global model of the terrestrial thermosphere, ionosphere, and protonosphere initially was solved by a transition to the constant 2o step on latitude for all parameters instead of the 5o step for parameters of the ionospheric F2 region and protonosphere and 10o step for parameters of the thermosphere and lower ionosphere, which had been used before. The version of the global model published by Barbolina et al.  and A. N. Namgaladze et al.  and developed for realization on personal computers was used as a basis for the transformation.
Transition to the 2o latitudinal net allowed numerical simulation of the quiet and disturbed ionosphere for a particular geophysical situation on March 24 and 25, 1987, comparison of the calculations with the EISCAT incoherent scatter data, and demonstration of quite satisfactory agreement between the simulation and observation results [ A. A. Namgaladze et al., 1993a, b, 1994]. However, the above transition required increase of the on-line and disk memory of the computer; so it was possible to install the model versions with the 2o step of latitudinal integration only at computers with the RAM memory of at least 16 Mb. A requirement of solution stability of the residual methods of the model equations also led to a reduction of the temporal integration steps down to 1-2 min instead of the 5 to 10-min steps used before. All these changes increased significantly the computer time needed for calculations.
High spatial resolution is needed only in the regions with high spatial gradient of the modeled parameters, in particular, in the auroral zones and adjoining regions; so it would be optimal to use a residual net with a variable step, decreasing where it is necessary. In ionospheric modeling, such algorithms are used in height integration. As for horizontal integration, we do not know any model using an inhomogeneous net. Apparently, the present results are the first in that direction.
Reconstructing the model to make computations by the inhomogeneous net on geomagnetic latitude, the transition to another net on any dimension was reduced to a change of several key parameters in one place of the initial program. Sufficient volumes of all the blocks in the program to work with the given net were provided automatically. At the beginning of the run, the program checks whether the data file is sufficient to work with the given net and whether the initial conditions were given at the same net. In such a way, complete protection from errors due to a transition from one residual net to another is provided.
The transition to an inhomogeneous latitude net consists of providing laws of changes of latitudinal steps of the "sphere" (the block for computing thermospheric parameters and molecular ions) and "tube" (the block for computing of parameters of the protonosphere and F2 region) as a function of the current latitude. The low is provided by the subprogram functions. To have a homogeneous latitudinal net (which is a particular case of the inhomogeneous net), these subprogram functions should return the same constant value for any current latitude. The same residual net is used for latitudinal integration in the block of the electric field potential as in the tube.
All the computer subprograms of the modeling part receive information on the used coordinate net in the form of special sets of data on knot latitudes for sphere and tubes. If one needs to determine the latitudinal step, for example, to calculate the derivatives, it is found as a residual between the latitudes of the adjoining knots. Owing to the inhomogeneity of the net, the steps from the right and from the left may differ, which leads to a considerable complication of the algorithms for numerical differentiation on latitude. The algorithms used allow fairly sharp changing of the net steps without losing the stability of the computation scheme. For example, in the net with a variable step, which was used for test computations (see below), the adjoining latitudinal steps differed by up to 30%.
Modifying the subprograms used to compute the electric field potentials and thermospheric parameters to work on the inhomogeneous latitudinal net, the subprograms were optimized with respect to the operation rate, which increased the total computation rate by about 10-15%.
The following obligatory requirements are imposed on the coordinate net in the new version of the model: the net knots for electric potential computation coincide with the bases of the geomagnetic field lines, which are the coordinate lines of the tube network; the longitudinal planes of the sphere and tube nets coincide; the longitudinal step is constant (in these computations it is equal to 15o), and the first longitudinal plane is situated at a longitude of 0o; the bases of the tube net lines and the points in which the potential is computed are situated at the altitude of the 16th knot of the sphere; the knots of the latitudinal net are symmetric in relation to the equator; the extreme latitudinal knots of the net are at the poles; and variation of the altitude step between two adjoining intervals does not exceed 10%. In other aspects the net may be arbitrary.
Test computations were performed to study the influence of the latitudinal step of the residual net on results of the model simulation. Computations by three versions of the program, which differ only in the coordinate net, were performed with the same temporal steps of 2 min for the sphere and tube from the same initial conditions interpolated from one net to another. The nets were (1) a homogeneous one with steps of 10o and 5o in the sphere and tube respectively (the "rough" net); (2) a homogeneous one with the same steps of 2o in the sphere and tube ("fine" net); and (3) an inhomogeneous net with the steps decreasing from 10o in the sphere and 5o in the tube at the equator to 2o in the auroral region.
Dependence of the latitudinal steps of integration on geomagnetic latitude in the sphere and tube is shown in Figure 1 (Figures 1b and 1a, respectively). It can be seen in the figure that the maximum of the net convergence (the step minimum) falls at a geomagnetic latitude of 70o. Figure 1 shows also the latitudinal knots of the net for all computations.
The computations were performed for the time intervals of 0000-0624 UT on March 24, 1987, with the same values of the model input parameters (the solar radiation flux, flux of the precipitating electrons, and field-aligned currents) as in the papers by A. A. Namgaladze et al. [1993a, 1993b, 1994] for quiet conditions. The results for 0624 UT and all three versions of the net are presented in Figures 2 and 3, where the computed latitudinal variations of the parameters of the northern hemisphere ionosphere and thermosphere are shown along the geomagnetic meridians of 210oand 30o, which correspond to the local time at the magnetic equator of 1546 LT and 0346 LT, respectively. The computed values in the net knots are shown by the open circles and the dashed line for the rough net, by the solid circles for the fine net, and by the points connected by the solid line for the inhomogeneous net.
In the top part of Figure 2 the computed latitudinal variations are presented for the electric field potential at an altitude of 175 km, above which the plasma is believed to be magnetized and the geomagnetic field lines are believed to be electrically equipotential. In the other parts of Figure 2 (from top to bottom) the computed latitudinal variations of the parameters of the ionospheric F2 region at 300 km are presented: logarithm of the electron concentration, electron temperature, and ion temperature. Figure 3 shows the computed latitudinal variations of the thermospheric parameters at 300 km (from top to bottom): temperature, logarithm of the atomic oxygen concentration, and the magnitude of the vector of the thermospheric wind horizontal velocity.
Figures 2 and 3 demonstrate that the strongest difference between the computed results for the rough and fine nets takes place in the values of the electric field potential and parameters of the ionospheric F2 region at high latitudes (above geomagnetic latitude of 60o). For example, the voltage across the polar cap in the rough net is 47 kV, and in the fine one it is only 18 kV (the top part of Figure 2) under the same values of the field-aligned currents, which are used as input parameters at the polar cap boundaries (geomagnetic latitude of 75o). The difference is due to the fact that the field-aligned currents decrease from their maximum value to zero at the distance of one latitudinal step of integration. This difference is 5o and 2o in the rough and fine nets, respectively (the width of the field-aligned current zone being less than the latitudinal step in the rough net); as a result, smaller electric charge is put into the ionosphere and the electric field value drops, respectively.
Owing to the reduced convection in the fine net, plasma transport is practically absent from the dayside through the pole to the nightside, which forms a so-called ionization "tongue." As a result, the electron concentrations in the near-pole F2 region are significantly lower in the fine net than in the rough one, and the electron temperatures are respectively higher (Figure 2). The ion temperature in the fine net is lower than in the rough one because Joule heating.
The differences in the thermospheric parameters (Figure 3) computed for the rough and fine nets are not concentrated only at high latitudes but are more or less homogeneously distributed around the globe. This is due to the fact that the horizontal transport of the charged components of the F2 region is important mainly at high latitudes, whereas the transport of the neutral thermospheric gas is equally important everywhere. The amplitude of the temperature variations at 300 km in computations with the fine net is 80 K less than that with the rough one. Correspondingly, the amplitude of the latitudinal variation of the atomic oxygen concentration and total density is less in the fine net. Lower horizontal gradients of the temperature lead to lower velocities of the horizontal winds. It is also worth noting the smoother latitudinal variations computed with the fine net compared to the rough one.
Thus, using the rough nets with the latitudinal integration steps exceeding 2o also leads, besides limitations related to the field-aligned currents and precipitations, to overestimation of the calculated electric fields and, accordingly, of the effects generated by them at high latitudes. Usage of the rough nets to investigate midlatitude phenomena is quite reasonable owing to their practicality.
Let us consider now the computations performed with the inhomogeneous net with variable (depending on latitude) step of the numerical integration of the model equation. Figure 1 shows that the inhomogeneous net is close to the fine one at geomagnetic latitudes above 60o and to the rough one at latitudes below 30o. The comparison of the model simulations for all three versions of the net is presented in Figures 2 and 3 and visually demonstrates that using the inhomogeneous net produces results for the majority of parameters close to those obtained with the fine net not only at high latitudes, but all around the globe.
It is essential that noticeable savings of computer resources is achieved without significant loss in accuracy. This fact allows performance of the model simulations at personal computers of the moderate class of the PC AT486 DX2 50, 66 type. On the whole, transition to the inhomogeneous net with variable integration step provides the following gains in comparison with the 2-degree (fine) net: about 3 times by the computation rate (and by 6 times if the temporal integration step is increased from 1 to 2 min), from 31 to 16 Mb in disk memory for the operating file of the model and from 7 to 3.8 Mb for saving the snapshot of the calculated parameter distribution, and from 14 to 7.6 Mb in on-line memory. As far as the stability of the computations is preserved, these facts demonstrate a high efficiency of the developed algorithms, which allow convergence of the net in the places where higher spatial resolution of the model is required.
Korenkov, Yu. N., et al., Comparison of global description of the ionospheric parameters in the theoretical and empirical ionospheric models, in Polar Ionosphere Studies, p. 16, PGI AN SSSR, Apatity, 1990.
Korenkov, Yu. N., et al., Model calculations of the ionospheric parameters for March 19, 1988, Geomagn. Aeron., 33 (1), 63, 1993.
Namgaladze, A. A., et al., Global model of the thermosphere-ionosphere-protonosphere system, Pure Appl. Geophys., 127 (2/3), 1988.
Namgaladze, A. A., et al., Global numerical model of the terrestrial thermosphere, ionosphere and protonosphere, Geomagn. Aeron., 30 (4), 612, 1990.
Namgaladze, A. A., et al., Numerical modeling of the thermosphere-ionosphere-protonosphere system, J. Atmos. Terr. Phys., 53 (11/12), 1113, 1991.
Namgaladze, A. A., M. A. Volkov, M. A. Martynenko, and A. N. Namgaladze, Numerical simulation of high-latitude effects of a substorm (abstract), in Mathematical Models of the Near Space, p. 39, NIIYaF MGU, Moscow, 1993a.
Namgaladze, A. A., et al., Numerical simulation of the ionospheric disturbance over EISCAT by the use of the global ionospheric model, in Sixth EISCAT Workshop, Andenes, Book of Abstracts, p. 17, 1993b.
Namgaladze, A. A., O. V. Martynenko, A. N. Namgaladze, and M. A. Volkov, Numerical simulation of the main ionospheric rough dynamics during the March 25, 1987 substorm, in Physics of the Auroral Phenomena, Preprint PGI-94-01-96, 22 pp., Polar Geophys. Inst., Apatity, 1994.
Namgaladze, A. N., O. V. Martynenko, V. V. Kachala, and E. V. Barbolina, Realization of the global numerical model of the terrestrial thermosphere, ionosphere and protonosphere at a personal computer (abstracts), in Mathematical Models of the Near Space, p. 82, NIIYaF MGU, Moscow, 1993.