The DTM-2000 empirical thermosphere model with new data assimilation and constraints at lower boundary: accuracy and properties

Abstract

The drag temperature model (DTM) is a semi-empirical model describing the temperature, density and composition of the Earth's thermosphere. Its first version (Ann. Geophys. 34 (1978) 9) used direct measurements of exospheric temperature and atmospheric densities derived from satellite drag data. It has later been refined (J. Geodesy 72 (1998) 161). However, both models have their lower boundaries at $\text{120}\text{km}$, which are not constrained by observations. Consequently in the lower thermosphere, the modelled temperature and density structure is uncertain. For predicting satellite orbits in the lower thermosphere, more realistic density models are required. We present a new DTM model having the following improvements:

• Temperature and its gradient at $\text{120}\text{km}$ are represented in agreement with theory and observation, using incoherent scatter radar and satellite-borne interferometer data.

• Atmosphere explorer (AE) data, which have not been assimilated in DTM-94, are used as they cover a complete solar activity cycle.

• The Mg II index is used whenever possible to represent the solar UV and EUV emission intensity instead of the solar decimetre radio flux, since it is more representative of solar instantaneous chromospheric activity than F_10.7 is.

The basic DTM mathematical representation of temperature and composition is used, with, however, some additions and modifications to take into account the variations at $\text{120}\text{km}$ altitude.

The temperature modelling accuracy has improved by 5–8%, and there is no model bias as a function of solar activity. The oxygen and helium modelling has improved as well, and this is demonstrated by the estimated drag scale coefficients issued from precise orbit computation. The scaling coefficients estimated using DTM-2000 are systematically closer to unity than those resulting from employing DTM-94 and MSIS-86 in the orbit computation. The minor constituents (O₂ and H) modelling is unchanged. The molecular nitrogen modelling is not improved, but this is, at least partly, caused by the poor data quality.

Despite these improvements, semi-empirical thermosphere models still suffer of weaknesses. First, they assume a steady-state equilibrium which is not necessarily reached in any circumstances. Second, the basic process of atmospheric heating by EUV is assumed to be represented by some indices (Mg II, F_10.7), while particle precipitation is represented by an index associated to a latitude without a longitude effect (K_p), although winds that transport energy suggest this effect. Third, the data accuracy and their incomplete geographical and temporal coverage are significant sources of uncertainty. Assimilation of a long time series of data, with a complete geographical coverage, and using the Mg II index will probably increase model accuracy from the present-day RMS of 19% to 10–15%. The complete CHAMP accelerometer data set may allow the achievement of that goal after 5 years of operations in 2005.

Introduction

Atmospheric density models are, besides their use in atmospheric studies, required to compute the atmospheric drag force in satellite orbit determination. They represent temperature and density as a function of altitude, latitude, local solar time, day of year, and parameters related to the state of atmospheric heating.

The accuracy of the model that is most used in precise orbit computations, DTM-94 (Berger et al., 1998), is about 20% (1σ) for its predicted total densities. The MSIS-86 model (Hedin, 1987) has comparable accuracy. DTM-94 is an improvement over the first Drag Temperature Model, DTM-78 (Barlier et al., 1978), which is not representative of low solar activity conditions due to the data set used in this model. This causes uncertain predictions under those conditions.

The inaccurate representation of temperature and density in the 120–$\text{220}\text{km}$ altitude range is due to the sparseness of data in this domain. The two basic reasons for the lower atmosphere modelling uncertainties are: first, the modelled density in the lower thermosphere results of adjustment of the model parameters at higher altitudes (220–$\text{1000}\text{km}$), which have been propagated downward by means of a modified Bates temperature profile (Bates, 1959); secondly, the modelling of the temperature and its gradient at $\text{120}\text{km}$ as constant values in DTM-94 causes uncertainties in the predicted density of the lower thermosphere, since significant temperature variability at that altitude is known from incoherent scatter radar observations (Alcaydé et al., 1974). However, this being acknowledged, the absence of precise data as well as data having enough spatial coverage, has lead to prefer constant boundary conditions.

This is why it is in this region that a special effort has been made to improve the modelling.

In this study, the temperature at $\text{120}\text{km}$ (T₁₂₀) is modelled using incoherent scatter radar observations. The latitudinal coverage is achieved by means of three incoherent scatter radar stations namely St. Santin (45°N), Arecibo (18°N) and the European Incoherent SCATter (EISCAT), which is at approximately 70°N, and is the only available data source at that latitude.

The WIND Imaging Interferometer (WINDII) is an instrument (Shepherd et al., 1993) placed onboard the Upper Atmosphere Research Satellite (UARS), which measured temperature and wind in the lower thermosphere. WINDII data are used to determine the vertical temperature gradients (dT₁₂₀) as a function of local time, latitude and season. They have also been used to derive relative density variations at $\text{120}\text{km}$ altitude (Bruinsma et al., 2002). The variations derived in that study are taken into account in DTM-2000.

Data from the Atmosphere Explorer-C and -E satellites (NSSDC CD-ROM) are assimilated in order to better reproduce in particular the temperature under low solar activity conditions.

The Mg II index (Heath and Schlesinger, 1986) is used as a proxy indicator to represent the solar activity in the EUV and UV whenever possible, which is to say, for data after 1978. Before that date, the solar radio flux at $\text{10.7}\text{cm}$ (F_10.7) is used. It has recently been demonstrated (Thuillier and Bruinsma, 2001) that the Mg II index is more representative of the solar radiation that governs upper atmosphere heating processes than F_10.7 is, while these indices are nearly 100% correlated over periods of at least several years. It is this property that enables both Mg II and F_10.7 to be used in the model adjustments.

The next section reviews the data used in this study, and presents the data processing that allowed the modelling of T₁₂₀ and dT₁₂₀. The DTM algorithms and the additional temperature and density variations (with respect to variations already represented in DTM-94) are described in Section 3. The results are presented in Section 4 which includes comparisons with DTM-94 and MSIS-86. The various results, the limiting factors in empirical modelling, the effects of data availability, and distribution and accuracy of the estimations, are discussed in Section 5. The future model improvements expected by new and consistent data are presented in Section 6.