中文摘要：

本文主要研究了利用GRACE星载加速度计数据反演热层密度.首先联合采用GRACE卫星2007—2009年星载加速度计数据和星载GPS数据进行动力学定轨并同时估计加速度计校正参数,依此对加速度计数据进行了校正.根据Sentman稀薄空气动力学方程计算卫星空气动力系数,对校正后的加速度计数据进行处理,反演得到了该时期沿轨热层大气密度.为分析反演密度的精度,将本文反演得到的GRACE-A卫星沿轨密度与Doornbos的解算结果,以及经验密度模型NRLMSISE00,HASDM模型进行比较分析.统计结果表明,本文反演结果比Doornbos系统性偏大约5%-8%,二者间的标准差（STD,standard deviation）在10%以内,具有较好的符合性.其差异主要是由于采用了不同的加速度标校以及空气动力系数计算策略.本文反演得到的热层密度较HASDM模型表现为正的系统性偏差且幅度在4%以内,而Doornbos的结果较HASDM模型约系统性偏小4%-7%,二者与HASDM模型的标准差均为30%左右.另外,本文反演密度与NRLMSISE00模型之间存在约30%-40%的系统性bias,其STD也在30%左右.

英文摘要：

This paper mainly discusses the thermosphere mass density derivation using the GRACE on-board accelerometer data.The precision of the accelerometer calibration as well as the satellite aerodynamics model have significant impact on the derivation.In this paper the accelerometer calibration is realized through precise orbit determination（POD）process by using the on-board GPS and accelerometer data,while the aerodynamic coefficients of the GRACE twin satellites are calculated using the Sentman equations.The GRACE accelerometer data during year2007—2009are processed and the densities along the satellites′orbit during this period areobtained.To assess the precision of the accelerometer-derived densities,densities from the GRACE-A satellite during 2007—2009are compared to the results from Doornbos（2012）,as well as the empirical NRLMSISE00 and HASDM models.The results indicate that the densities from this paper are systematically about 8%larger than those from Doornbos（2012）,while the STD between these two independent dataset is under 10%.Their differences are mainly attributed to the different strategies adopted for accelerometer calibration and aerodynamic coefficients calculation.The accelerometer-derived densities in this paper are systematically larger than the HASDM model with a bias about 4%,while the densities from Doornbos are 4%-7%lower.The STDs of the derived densities and the results of Doornbos with respect to HASDM model are both around 30%.The bias and STD between the accelerometer-derived densities and the empirical NRLMSISE00 model are 30%-40% and around 30%,respectively.