Precise orbit determination using the batch filter based on particle filtering with genetic resampling approach

In this study, genetic resampling (GRS) approach is utilized for precise orbit determination (POD) using the batch filter based on particle filtering (PF). Two genetic operations, which are arithmetic crossover and residual mutation, are used for GRS of the batch filter based on PF (PF batch filter). For POD, Laser-ranging Precise Orbit Determination System (LPODS) and satellite laser ranging (SLR) observations of the CHAMP satellite are used. Monte Carlo trials for POD are performed by one hundred times. The characteristics of the POD results by PF batch filter with GRS are compared with those of a PF batch filter with minimum residual resampling (MRRS). The post-fit residual, 3D error by external orbit comparison, and POD repeatability are analyzed for orbit quality assessments. The POD results are externally checked by NASA JPL’s orbits using totally different software, measurements, and techniques. For post-fit residuals and 3D errors, both MRRS and GRS give accurate estimation results whose mean root mean square (RMS) values are at a level of 5 cm and 10–13 cm, respectively. The mean radial orbit errors of both methods are at a level of 5 cm. For POD repeatability represented as the standard deviations of post-fit residuals and 3D errors by repetitive PODs, however, GRS yields 25% and 13% more robust estimation results than MRRS for post-fit residual and 3D error, respectively. This study shows that PF batch filter with GRS approach using genetic operations is superior to PF batch filter with MRRS in terms of robustness in POD with SLR observations.     

Precision orbit determination performance for CryoSat-2

In this paper we discuss our efforts to perform precision orbit determination (POD) of CryoSat-2 which depends on Doppler and satellite laser ranging tracking data. A dynamic orbit model is set-up and the residuals between the model and the tracking data is evaluated. The average r.m.s. of the 10?s averaged Doppler tracking pass residuals is approximately 0.39?mm/s; and the average of the laser tracking pass residuals becomes 1.42?cm. There are a number of other tests to verify the quality of the orbit solution, we compare our computed orbits against three independent external trajectories provided by the CNES. The CNES products are part of the CryoSat-2 products distributed by ESA. The radial differences of our solution relative to the CNES precision orbits shows an average r.m.s. of 1.25?cm between Jun-2010 and Apr-2017. The SIRAL altimeter crossover difference statistics demonstrate that the quality of our orbit solution is comparable to that of the POE solution computed by the CNES. In this paper we will discuss three important changes in our POD activities that have brought the orbit performance to this level. The improvements concern the way we implement temporal gravity accelerations observed by GRACE; the implementation of ITRF2014 coordinates and velocities for the DORIS beacons and the SLR tracking sites. We also discuss an adjustment of the SLR retroreflector position within the satellite reference frame. An unexpected result is that we find a systematic difference between the median of the 10 s Doppler tracking residuals which displays a statistically significant pattern in the South Atlantic Anomaly (SSA) area where the median of the velocity residuals varies in the range of ?0.15 to +0.15?mm/s.     

Characteristics of GOCE orbits based on Satellite Laser Ranging

The Gravity field and steady-state Ocean Circulation Explorer (GOCE) was the first European Space Agency’s (ESA) Earth Explorer core mission. Through its extremely low, about 260?km above the Earth, circular, sun-synchronous orbit, the satellite gained high spatial resolution and accuracy gravity gradient, and ocean circulation data. Global Positioning System (GPS) receivers, mounted on the spacecraft, allowed the determination of reduced-dynamic and kinematic GOCE orbits, whereas Laser Retroreflector Array (LRA) dedicated to Satellite Laser Ranging (SLR) allowed an independent validation of GPS-derived orbits. In this paper, residuals between different GPS-based orbit types and SLR observations are used to investigate the sensitivity and the influence of solar, geomagnetic, and ionospheric activities on the quality of kinematic and reduced-dynamic GOCE orbits. We also analyze the quality of data provided by individual SLR sites, by detecting time biases using ascending and descending sun-synchronous GOCE orbit passes, and the residual analysis of the measurement characteristics, i.e., the dependency of SLR residuals as a function of nadir and horizontal angles. Results show a substantial vulnerability of kinematic orbit solutions to the solar F10.7 index and the ionospheric activity measured by the variations of the Total Electron Content (TEC) values. The sensitivity of kinematic orbits to the three-hour-range KP index is rather minor. The reduced-dynamic orbits are almost insensitive to indices describing ionospheric, solar, and geomagnetic activities. The investigation of individual SLR sites shows that some of them are affected by time bias errors, whereas other demonstrate systematics, such as a dependency between observation residuals and the satellite nadir angle or the horizontal azimuth angle from the SLR station to the direction of the satellite.     

BeiDou-3 orbit and clock quality of the IGS Multi-GNSS Pilot Project

Within the Multi-GNSS Pilot Project (MGEX) of the International GNSS Service (IGS), precise orbit and clock products for the BeiDou-3 global navigation satellite system (BDS-3) are routinely generated by a total of five analysis centers. The processing standards and specific properties of the individual products are reviewed and the BDS-3 orbit and clock product performance is assessed through direct inter-comparison, satellite laser ranging (SLR) residuals, clock stability analysis, and precise point positioning solutions. The orbit consistency evaluated by the signal-in-space range error is on the level of 4–8 cm for the medium Earth orbit satellites whereas SLR residuals have RMS values between 3 and 9 cm. The clock analysis reveals sytematic effects related to the elevation of the Sun above the orbital plane for all ACs pointing to deficiencies in solar radiation pressure modeling. Nevertheless, precise point positioning with the BDS-3 MGEX orbit and clock products results in 3D RMS values between 7 and 8 mm.     

The effect of geocenter motion on Jason-2 orbits and the mean sea level

We compute a series of Jason-2 GPS and SLR/DORIS-based orbits using ITRF2005 and the std0905 standards ( Lemoine et al., 2010). Our GPS and SLR/DORIS orbit data sets span a period of 2 years from cycle 3 (July 2008) to cycle 74 (July 2010). We extract the Jason-2 orbit frame translational parameters per cycle by the means of a Helmert transformation between a set of reference orbits and a set of test orbits. We compare the annual terms of these time-series to the annual terms of two different geocenter motion models where biases and trends have been removed. Subsequently, we include the annual terms of the modeled geocenter motion as a degree-1 loading displacement correction to the GPS and SLR/DORIS tracking network of the POD process. Although the annual geocenter motion correction would reflect a stationary signal in time, under ideal conditions, the whole geocenter motion is a non-stationary process that includes secular trends. Our results suggest that our GSFC Jason-2 GPS-based orbits are closely tied to the center of mass (CM) of the Earth consistent with our current force modeling, whereas GSFC’s SLR/DORIS-based orbits are tied to the origin of ITRF2005, which is the center of figure (CF) for sub-secular scales. We quantify the GPS and SLR/DORIS orbit centering and how this impacts the orbit radial error over the globe, which is assimilated into mean sea level (MSL) error, from the omission of the annual term of the geocenter correction. We find that for the SLR/DORIS std0905 orbits, currently used by the oceanographic community, only the negligence of the annual term of the geocenter motion correction results in a – 4.67 ± 3.40 mm error in the Z-component of the orbit frame which creates 1.06 ± 2.66 mm of systematic error in the MSL estimates, mainly due to the uneven distribution of the oceans between the North and South hemisphere.     

Impact of ITRS 2014 realizations on altimeter satellite precise orbit determination

This paper evaluates orbit accuracy and systematic error for altimeter satellite precise orbit determination on TOPEX, Jason-1, Jason-2 and Jason-3 by comparing the use of four SLR/DORIS station complements from the International Terrestrial Reference System (ITRS) 2014 realizations with those based on ITRF2008. The new Terrestrial Reference Frame 2014 (TRF2014) station complements include ITRS realizations from the Institut National de l’Information Géographique et Forestière (IGN) ITRF2014, the Jet Propulsion Laboratory (JPL) JTRF2014, the Deutsche Geodätisches Forschungsinstitut (DGFI) DTRF2014, and the DORIS extension to ITRF2014 for Precise Orbit Determination, DPOD2014. The largest source of error stems from ITRF2008 station position extrapolation past the 2009 solution end time. The TRF2014 SLR/DORIS complement impact on the ITRF2008 orbit is only 1–2 mm RMS radial difference between 1992–2009, and increases after 2009, up to 5 mm RMS radial difference in 2016. Residual analysis shows that station position extrapolation error past the solution span becomes evident even after two years, and will contribute to about 3–4 mm radial orbit error after seven years. Crossover data show the DTRF2014 orbits are the most accurate for the TOPEX and Jason-2 test periods, and the JTRF2014 orbits for the Jason-1 period. However for the 2016 Jason-3 test period only the DPOD2014-based orbits show a strong and statistically significant margin of improvement. The positive results with DTRF2014 suggest the new approach to correct station positions or normal equations for non-tidal loading before combination is beneficial. We did not find any compelling POD advantage in using non-linear over linear station velocity models in our SLR & DORIS orbit tests on the Jason satellites. The JTRF2014 proof-of-concept ITRS realization demonstrates the need for improved SLR+DORIS orbit centering when compared to the Ries (2013) CM annual model. Orbit centering error is seen as an annual radial signal of 0.4 mm amplitude with the CM model. The unmodeled CM signals show roughly a 1.8 mm peak-to-peak annual variation in the orbit radial component. We find the TRF network stability pertinent to POD can be defined only by examination of the orbit-specific tracking network time series. Drift stability between the ITRF2008 and the other TRF2014-based orbits is very high, the relative mean radial drift error over water is no larger than 0.04 mm/year over 1993–2015. Analyses also show TRF induced orbit error meets current altimeter rate accuracy goals for global and regional sea level estimation.     

基于单频星载GPS数据的低轨卫星精密定轨

为满足搭载单频GPS接收机低轨卫星的精密定轨需求以及深化单频定轨研究,文中解决了单频星载GPS数据的周跳探测问题,并利用“海洋二号”（HY-2A）卫星及“资源三号”（ZY-3）卫星的单频星载GPS实测数据采用两种方法确定了二者的简化动力学轨道,并通过观测值残差分析、与双频精密轨道比较、激光测卫数据检核等方法对所得轨道精度进行评定。结果表明,在不考虑电离层延迟影响的情况下,HY-2A卫星定轨精度为2~3dm,ZY-3卫星为1m左右;而采用半和改正组合消除电离层延迟一阶项影响后,二者定轨精度均显著提高,HY-2A卫星三维精度提高至1dm左右,ZY-3卫星提高至1~2dm。文章的研究成果表明,搭载单频GPS接收机的低轨卫星也可获得厘米级的定轨精度。     

Impact of GPS antenna phase center variations on precise orbits of the GOCE satellite

The first European Space Agency Earth explorer core mission GOCE (Gravity field and steady-state Ocean Circulation Explorer) has been launched on March 17, 2009. The 12-channel dual-frequency Global Positioning System receiver delivers 1 Hz data and provides the basis for precise orbit determination (POD) on the few cm-level for such a very low orbiting satellite (254.9 km). As a member of the European GOCE Gravity Consortium, which is responsible for the GOCE High-level Processing Facility (HPF), the Astronomical Institute of the University of Bern (AIUB) provides the Precise Science Orbit (PSO) product for the GOCE satellite. The mission requirement for 1-dimensional POD accuracy is 2 cm. The use of in-flight determined antenna phase center variations (PCVs) is necessary to meet this requirement. The PCVs are determined from 154 days of data and the magnitude is up to 3-4 cm. The impact of the PCVs on the orbit determination is significant. The cross-track direction benefits most of the PCVs. The improvement is clearly seen in the orbit overlap analysis and in the validation with independent Satellite Laser Ranging (SLR) measurements. It is the first time that SLR could validate the cross-track component of a LEO orbit.     

CHAMP and GRACE accelerometer calibration by GPS-based orbit determination

Current and planned Earth observation missions are equipped with highly sensitive accelerometers. Before using the data, the instrument has to be calibrated by determining scale and bias parameters for each axis. Here, the accelerometer measurements are used in a GPS-based reduced-dynamic orbit determination approach, replacing the non-gravitational force models, and nominally daily calibration parameters are estimated. Additional empirical accelerations are estimated to account for deficiencies in the applied force models. This method is applied to 5 years of CHAMP and GRACE data, resulting in an orbit precision at the level of a few centimeters. In along-track direction the calibration parameters can be estimated freely, scale factors of 0.96 ± 0.014 and 0.95 ± 0.015 are obtained for GRACE A and B, and 0.85 ± 0.024 for CHAMP. A constant scale factor results in the smoothest bias series, with clear trends and occasional jumps. In radial and cross-track direction tight constraints to a priori biases have to be applied. Furthermore, the determined orbits are analyzed with respect to reference trajectories, and SLR, phase and KBR residuals are presented.     

Generating precise and homogeneous orbits for Jason-1 and Jason-2

Driven by the GMES (Global Monitoring for Environment and Security) and GGOS (Global Geodetic Observing System) initiatives the user community has a strong demand for high-quality altimetry products. In order to derive such high-quality altimetry products, precise orbits for the altimetry satellites are a necessity. With the launch of the TOPEX/Poseidon mission in 1992 a still on-going time series of high-accuracy altimetry measurements of ocean topography started, continued by the altimetry missions Jason-1 in 2001 and Jason-2/OSTM in 2008. This paper contributes to the on-going orbit reprocessing carried out by several groups and presents the efforts of the Navigation Support Office at ESA/ESOC using its NAPEOS software for the generation of precise and homogeneous orbits referring to the same reference frame for the altimetry satellites Jason-1 and Jason-2. Data of all three tracking instruments on-board the satellites (beside the altimeter), i.e. GPS, DORIS, and SLR measurements, were used in a combined data analysis. About 7 years of Jason-1 data and more than 1 year of Jason-2 data were processed. Our processing strategy is close to the GDR-C standards. However, we estimated slightly different scaling factors for the solar radiation pressure model of 0.96 and 0.98 for Jason-1 and Jason-2, respectively. We used 30 s sampled GPS data and introduced 30 s satellite clocks stemming from ESOC’s reprocessing of the combined GPS/GLONASS IGS solution. We present the orbit determination results, focusing on the benefits of adding GPS data to the solution. The fully combined solution was found to give the best orbit results. We reach a post-fit RMS of the GPS phase observation residuals of 6 mm for Jason-1 and 7 mm for Jason-2. The DORIS post-fit residuals clearly benefit from using GPS data in addition, as the DORIS data editing improves. The DORIS observation RMS for the fully combined solution is with 3.5 mm and 3.4 mm, respectively, 0.3 mm better than for the DORIS-SLR solution. Our orbit solution agrees well with external solutions from other analysis centers, as CNES, LCA, and JPL. The orbit differences between our fully combined orbits and the CNES GDR-C orbits are of about 0.8 cm for Jason-1 and at 0.9 cm for Jason-2 in the radial direction. In the cross-track component we observe a clear improvement when adding GPS data to the POD process. The 3D-RMS of the orbit differences reveals a good orbit consistency at 2.7 cm and 2.9 cm for Jason-1 and Jason-2. Our resulting orbit series for both Jason satellites refer to the ITRF2005 reference frame and are provided in sp3 file format on our ftp server.     

Dynamic and non-dynamic ERS-1 radial orbit improvement from ERS-1/TOPEX dual-satellite altimetry

Dual-satellite altimeter crossover differences between ERS-1 and TOPEX/Poseidon have been included as supplementary tracking data in ERS-1 orbit computations from SLR and single-satellite crossover differences. It was found that including the dual-satellite crossover differences slightly improves the ERS-1 radial orbit accuracy of about 12 cm for orbits computed with the JGM-2 gravity field and also leads to a better ‘centering’ of the ERS-1 orbit in the terrestrial reference frame defined for TOPEX/Poseidon. In addition to this dynamic orbit improvement technique, a non-dynamic technique has been investigated that removes the larger part of the ERS-1 radial orbit error from the dual-satellite crossover difference residuals. For ERS-1 orbits computed with the GEM-T2 gravity field, it was found that the non-dynamic technique could improve the radial orbit accuracy from 140 cm to the same level of accuracy as the ERS-1 JGM-2 orbits.     

Achievable debris orbit prediction accuracy using laser ranging data from a single station

Earlier studies have shown that an orbit prediction accuracy of 20 arc sec ground station pointing error for 1–2 day predictions was achievable for low Earth orbit (LEO) debris using two passes of debris laser ranging (DLR) data from a single station, separated by about 24 h. The accuracy was determined by comparing the predicted orbits with subsequent tracking data from the same station. This accuracy statement might be over-optimistic for other parts of orbit far away from the station. This paper presents the achievable orbit prediction accuracy using satellite laser ranging (SLR) data of Starlette and Larets under a similar data scenario as that of DLR. The SLR data is corrupted with random errors of 1 m standard deviation so that its accuracy is similar to that of DLR data. The accurate ILRS Consolidated Prediction Format orbits are used as reference to compute the orbit prediction errors. The study demonstrates that accuracy of 20 arc sec for 1–2 day predictions is achievable.     

Towards development of a consistent orbit series for TOPEX,Jason-1, and Jason-2

The TOPEX/Poseidon, Jason-1 and Jason-2 set of altimeter data now provide a time series of synoptic observations of the ocean that span nearly 17 years from the launch of TOPEX in 1992. The analysis of the altimeter data including the use of altimetry to monitor the global change in mean sea level requires a stable, accurate, and consistent orbit reference over the entire time span. In this paper, we describe the recomputation of a time series of orbits that rely on a consistent set of reference frames and geophysical models. The recomputed orbits adhere to the IERS 2003 standards for ocean and earth tides, use updates to the ITRF2005 reference frame for both the SLR and DORIS stations, apply GRACE-derived models for modeling of the static and time-variable gravity, implement the University College London (UCL) radiation pressure model for Jason-1, use improved troposphere modeling for the DORIS data, and apply the GOT4.7 ocean tide model for both dynamical ocean tide modeling and for ocean loading. The new TOPEX orbits have a mean SLR fit of 1.79 cm compared to 2.21 cm for the MGDR-B orbits. These new TOPEX orbits agree radially with independent SLR/crossover orbits at 0.70 cm RMS, and the orbit accuracy is estimated at 1.5–2.0 cm RMS over the entire TOPEX time series. The recomputed Jason-1 orbits agree radially with the Jason-1 GDR-C orbits at 1.08 cm RMS. The GSFC SLR/DORIS dynamic and reduced-dynamic orbits for Jason-2 agree radially with independent orbits from the CNES and JPL at 0.70–1.06 cm RMS. Applying these new orbits, and using the latest altimeter corrections for TOPEX, Jason-1, and Jason-2 from September 1992 to May 2009, we find a global rate in mean sea level of 3.0 ± 0.4 mm/yr.     

附加先验轨道约束的LEO几何法实时精密定轨验证分析

低轨卫星的实时精密定轨能够极大拓展其完成复杂科学任务的能力,例如实时环境监测、机动控制和卫星自主导航等.本文根据几何法实时精密定轨模型,提出了附加LEO先验轨道约束从而改善实时定轨的精度、收敛速度和稳定性的构想.分别采用广播星历、超快速星历预报部分和实时精密星历,设计了6种实时定轨方案,并利用Swarm-A,B,C星7天的观测数据进行方案验证与分析.结果表明,使用广播星历、IGU和IGC星历的方案精度递增,附加先验轨道约束能够进一步提升精度.使用IGC星历并附加标准差为1m的先验轨道约束后,在径向、切向和法向的定轨精度分别达到6.12cm,5.55cm和4.98cm.此外,附加先验轨道约束能够显著提升收敛速度,使用IGC星历平均收敛时间约为31min,附加标准差为1m的先验轨道约束后收敛仅需约4min.      

Assessing attitude error of FORMOSAT-3/COSMIC satellites and its impact on orbit determination

An attitude determination and control system (ADCS) is critical to satellite attitude maneuvers and to the coordinate transformation from the inertial frame to the spacecraft frame. This paper shows specific sensors in the ADCS of the satellite mission FORMOSAT-3/COSMIC (F3/C) and the impact of the ADCS quality on orbit accuracy. The selection of main POD antenna depends on the beta angles of the different F3/C satellites (for FM2 and FM4) during the inflight phase. In particular, under the eclipse, alternative attitude sensors are activated to replace the Sun sensors, and such a sensor change leads to anomalous GPS phase residuals and a degraded orbit accuracy. Since the nominal attitude serves as a reference for ADCS, the 3-dimensional attitude-induced errors in reduced dynamic orbits over selected days in 2010 show 9.35, 10.78, 4.97, 5.48, 7.18, and 6.89 cm for FM1–FM6. Besides, the 3-dimensional velocity errors induced by the attitude effect are 0.10, 0.10, 0.07, 0.08, 0.09, and 0.10 for FM1–FM6. We analyze the quality of the observed attitude transformation matrix of F3/C and its impact on kinematic orbit determination. With 249 days of GPS in 2008, the analysis leads to the following averaged 3-dimensional attitude-induced orbit errors: 2.72, 2.62, 2.37, 1.90, 1.70, and 1.99 cm for satellites FM1–FM6. Critical suggestions of geodetic payloads for the follow-on mission of F3/C are presented based on the current result.     

Geosynchronous satellites expanding a future GNSS satellite constellation: A precise orbit determination study

The integration of geosynchronous orbit (GSO) satellites in Global Navigation Satellite Systems (GNSS) is mostly discussed to enable a regional enhancement for tracking. But how do GSO satellites affect the orbit determination of the rest of the constellation? How accurately can these orbits be determined in a future GNSS tracking scenario with optical links? In this simulation study we analyze the benefit of GSO satellites as an expansion of a MEO (Medium Earth Orbit) satellite constellation – we selected the Galileo satellite constellation – for MEO Precise Orbit Determination (POD). We address not only the impact on POD of MEO satellites but also the possibility to precisely determine the GSO satellites – geostationary orbits (GEO) and inclined geosynchronous orbits (IGSO) – in such an expanded MEO constellation. In addition to GNSS microwave observations, we analyze the influence of different optical links between the participating entities: Optical two-way Inter-Satellite Links (OISL) and ground-space oriented Optical Two-Way Links (OTWL). These optical measurements together with the GNSS microwave observations give a remarkable benefit for the POD capability. In the case of GNSS and OTWL, we simulate the measurements with regard to a network of 16 ground stations. We pay great attention to the simulation of systematic effects of all measurement techniques. We discuss the influence on the systematic errors as well as the formal orbit uncertainties. A MEO constellation expanded with GSO satellites as well as the use of optical links together with GNSS observations not only improves the MEO satellite orbits but also the GSOs to a great extent.     

Comparison of electron density measurements from CSES and Swarm satellites with GNSS ionospheric tomography data

Electron density measurements obtained from China Seismo‐Electromagnetic Satellite (CSES) and Swarm-B can play an increasingly important role in the study of ionosphere above F2 peak height. This study presented a comprehensive comparison of electron density products obtained from Langmuir probe mounted on CSES and Swarm-B with ionospheric tomography for a whole year period of 2019. CSES was fully compared with Swarm-B on a global scale, including both absolute and relative differences, and a new index called NFI was developed to better quantify the similarity between two latitudinal profiles of electron density. CSES and Swarm-B were then compared with tomography respectively in four regions, roughly located in America, Europe, Australia and China. Results indicated that CSES data are consistent with Swarm-B, as NFI values exceed 0.6 for most of the analyses. Tomography and Swarm-B were found to have a good agreement as their biases are less than 0.2 × 10⁵ el/cm³ in general. For the comparison between CSES and tomography, the bias increased to around 0.6 × 10⁵ el/cm³ but the standard deviation changed slightly, validating the underestimation of electron density by CSES. The spatiotemporal comparisons of CSES and Swarm-B with tomography showed that: 1) the differences in electron density were relatively low in middle latitudes and increased rapidly in the regions of equatorial ionization anomaly; 2) Swarm-B has a better consistent with tomography than CSES, but both are capable of detecting ionosphere anomalies such as midlatitude arcs; and 3) CSES and Swarm-B both can capture the seasonal changes of electron density, while their values are basically smaller than those from tomography in Spring and Summer months.     

Thermosphere densities derived from Swarm GPS observations

After the detection of many anomalies in the Swarm accelerometer data, an alternative method has been developed to determine thermospheric densities for the three-satellite mission. Using a precise orbit determination approach, non-gravitational and aerodynamic-only accelerations are estimated from the high-quality Swarm GPS data. The GPS-derived non-gravitational accelerations serve as a baseline for the correction of the Swarm-C along-track accelerometer data. The aerodynamic accelerations are converted directly into thermospheric densities for all Swarm satellites, albeit at a much lower temporal resolution than the accelerometers would have been able to deliver. The resulting density and acceleration data sets are part of the European Space Agency Level 2 Swarm products.To improve the Swarm densities, two modifications have recently been added to our original processing scheme. They consist of a more refined handling of radiation pressure accelerations and the use of a high-fidelity satellite geometry and improved aerodynamic model. These modifications lead to a better agreement between estimated Swarm densities and NRLMSISE-00 model densities. The GPS-derived Swarm densities show variations due to solar and geomagnetic activity, as well as seasonal, latitudinal and diurnal variations. For low solar activity, however, the aerodynamic signal experienced by the Swarm satellites is very small, and therefore it is more difficult to accurately resolve latitudinal density variability using GPS data, especially for the higher-flying Swarm-B satellite. Therefore, mean orbit densities are also included in the Swarm density product.