Demonstrating developments in highfidelity analytical radiation force modelling methods for spacecraft with a new model for GPS IIR/IIRM
Abstract
This paper presents recently developed strategies for highfidelity, analytical radiation force modelling for spacecraft. The performance of these modelling strategies is assessed using a new model for the Global Positioning System Block IIR and IIRM spacecraft. The statistics of various orbit model parameters in a full orbit estimation process that uses tracking data from 100 stations are examined. Over the full year of 2016, considering all Block IIR and IIRM satellites on orbit, introducing University College London’s gridbased model into the orbit determination process reduces mean 3d orbit overlap values by 9% and the noise about the mean orbit overlap value by 4%, when comparing against orbits estimated using a simpler boxwing model of the spacecraft. Comparing with orbits produced using the extended Empirical CODE Orbit Model, we see decreases of 4% and 3% in the mean and the noise about the mean of the 3d orbit overlap statistics, respectively. In orbit predictions over 14day intervals, over the first day, we see smaller rootmeansquare errors in the alongtrack and crosstrack directions, but slightly larger errors in the radial direction. Over the 14th day, we see smaller errors in the radial and crosstrack directions, but slightly larger errors in the alongtrack direction.
Keywords
Solar radiation pressure Analytical force models GPS Orbit determination Orbit prediction1 Introduction
At Global Navigation Satellite System (GNSS) altitudes, the major force that affects satellite trajectories is Earth gravity, but the effects of solar and lunar gravity are also considerable. Next in this hierarchy of forces, as shown in Fig. 3.1 of Montenbruck and Gill (2000), there is solar radiation pressure (SRP), which is caused by the interaction of solar photons with spacecraft surfaces. If this is not accounted for in the force models used for orbit prediction, the calculated position of the spacecraft can be in error at the 100m level after 12 h (Fliegel and Gallini 1996). For this reason, SRP modelling is an important topic in GNSS orbit determination and precise orbit determination (POD) in general (Montenbruck and Gill 2000; Vallado and McClain 2001; Tapley et al. 2004).
Since the 1980s, various methods for dealing with the problem have been presented in the literature (Colombo 1986; Beutler et al. 1994; Fliegel and Gallini 1996; Springer et al. 1999; BarSever and Kuang 2003; Arnold et al. 2015). Many of these are empirical methods, requiring no a priori knowledge of the spacecraft properties or its operating environment. In global network analyses that incorporate tracking measurements from a large network of one hundred stations or more, such methods can produce spacecraft orbits with cmlevel accuracy (Sośnica et al. 2015).
However, in a purely empirical approach, the orbit model parameters can absorb the effects of other unmodelled or mismodelled processes (e.g. Earth rotation, geocentre variation (Meindl et al. 2013), etc.). This can result in orbit model parameter estimates that are nonphysical, which means they cannot improve our understanding of the physical processes that determine the trajectory of the satellites and are therefore limited in their ability to help improve the modelling of those processes. As a result, a number of groups introduced analytical, or physicsbased, radiation force modelling into their orbit estimation processes. In this area, the boxwing (BW) approach, first introduced by Marshall and Luthcke (1994) for application to POD of the TOPEX/Poseidon mission, has been particularly influential. The general concept is to model the spacecraft structure using eight flat plates (six for a cuboid representing the spacecraft bus and two for solar panels), with assumed values for the optical and thermal properties of the surfaces, which are then combined with a priori modelling of the spacecraft attitude and the incident radiation fluxes. This approach was applied to the Block II/IIA and Block IIR satellites of the Global Positioning System (GPS) by RodriguezSolano et al. (2012). In comparing the performance of their semianalytical adjustable boxwing model with the Centre for Orbit Determination (CODE) Empirical Orbit Model (ECOM; Beutler et al. 1994), the authors determined that the orbit solutions produced by the two methods were comparable, but that the accelerations produced by the ECOM model were less physically meaningful. More recent efforts that adopt a broadly similar modelling approach include Montenbruck et al. (2015) for Galileo satellites and Montenbruck et al. (2017a) for the QZS1 satellite of QuasiZenith Satellite System (QZSS).
The boxwing models are relatively easy to implement. However, they are not able to fully capture the radiation fluxspacecraft surface interaction in satellites with complex surface geometry where the effects of mutual selfshadowing and reflected radiation can be significant. An alternative class of analytical radiation force modelling methods, in which raytracing techniques are combined with detailed spacecraft surface models to account for SRP, Earth radiation pressure (ERP) and thermal reradiation (TRR), was developed in the early 1990s (Klinkrad et al. 1991), and these methods are able to capture these detailed effects. The models were tested in POD of the European remote sensing satellites ERS2 and ENVISAT (Doornbos et al. 2002). To distinguish them from the boxwing methods, we refer to these as highfidelity analytical radiation force modelling methods. In GNSS, highfidelity SRP modelling for GLONASS satellites was first explored by Ziebart and Dare (2001). This work was motivated by broader efforts to improve GLONASS orbit quality as part of the IGEX98 campaign (Willis et al. 1999). Work in this area continued over the years at University College London (UCL), where the approach was enhanced with methods to account for TRR (Adhya 2005), ERP (Sibthorpe 2006; Ziebart et al. 2007; Li et al. 2017) and antenna thrust (AT; Ziebart et al. 2007), and validated on a number of additional cases including the GPS Block IIA and Block IIR satellites, the Jason1 spacecraft of the Ocean Surface Topography Mission and ENVISAT (Ziebart et al. 2005; Sibthorpe 2006). Recent work in this area demonstrated improved accuracy in shadow modelling when using geometric primitives, as opposed to triangular tessellations, to represent curved surfaces when constructing the spacecraft model (Grey and Ziebart 2014). The modelling approach, as presented in Ziebart et al. (2005), was adopted into the operational standards for precise orbit determination of the Jason1 altimetry satellite (Cerri et al. 2010; Zelensky et al. 2010). Recently, other research groups have explored a broadly similar approach for modelling SRP on Beidou satellites (Tan et al. 2016; Wang et al. 2018), on the Gravity Field and SteadyState Ocean Circulation Explorer (GOCE) satellite (Gini 2014) and on QZS1 (Darugna et al. 2018).
In this paper, using a new model for the GPS Block IIR and IIRM satellites, we present developments to the UCL model computation strategies that are designed to extend the validity of the models to all possible orientations of the radiation source(s) with respect to the spacecraft. As such, these nextgeneration generalpurpose models account for the effect of radiation forcing from any number of radiation sources, from any direction, providing a highfidelity radiation flux–spacecraft interaction model that can be used to deal with both SRP and ERP. A key advantage of this approach is that the final model makes no prior assumptions about the attitude characteristics of the spacecraft and can therefore deal with any deviations from nominal attitude.
2 UCL modelling strategy
 (i)
Computation of the bus model, where the space vehicle bus contribution to the accelerations due to SRP and ERP is dealt with. In this process, the accelerations due to thermal emissions from the multilayered insulation (MLI) covering the bus are also computed, according to Sect. 4.3.3 of Adhya (2005). The core technique uses a raytracing algorithm, where the rays simulate the incident radiation flux for a given geometry of the spacecraft with respect to the radiation source. The output of this process is a set of three grids representing the accelerations in the X, Y and Zaxes of the spacecraft bodyfixed system (BFS), where the grid nodes are spaced at 1° intervals in latitude and longitude in the BFS.
 (ii)
Separate computation of the solar panel model, where the solar panel contributions to the accelerations due to SRP and ERP are dealt with.
 (iii)
AT modelling, which accounts for the recoil force on the spacecraft due to emission of photons from signal transmitters.
As input, the approach requires a computer model of the spacecraft that holds information about the external geometry and various surface material properties including reflectivity, specularity, absorptivity and emissivity. The models are built from a combination of geometric primitives (polygons, circles, cylinders, spheres, cones and truncated cones), avoiding any need for tessellation, especially on curved surfaces (Ziebart et al. 2003). This produces models with good geometric fidelity without requiring an excessively large number of components, e.g. the UCL model for the GPS IIR/IIRM bus, as shown in Fig. 6, is made up of 182 components. The solar panels (not shown in Fig. 6) are modelled as two rectangular plates.
For solar flux, the models are computed using a nominal value for the mean solar irradiance at one astronomical unit (AU) of 1368 Wm^{−2} (Hastings and Garrett 1996). The solar irradiance is known to vary over the solar cycle (with a period of between 9 and 14 years) by 1.4 Wm^{−2}. This represents circa 0.1% variation in the parameter. Little is gained by correcting the nominal value. It is more important to scale the model depending on the probeSun distance at the calculation epoch. Taking 1368 Wm^{−2} as a reference value, the eccentricity of the Earth’s orbit about the Sun modulates the solar irradiance near the Earth to 1415.7 Wm^{−2} at perihelion (+3.4%) and 1322.6 Wm^{−2} at aphelion (−3.3%). This gives a variation (between perihelion and aphelion) of circa 100 Wm^{−2} (the precise value being 91.3 Wm^{−2}), approximately 6.7% of the mean value.
For the Earth radiation flux model, we use data from the Clouds and Earth Radiant Energy System (CERES) (Wielicki et al. 1996) project, which provides the irradiance at the topofatmosphere (TOA), an altitude of ~ 30 km above the Earth’s surface, in a grid format spaced at 1degree intervals in latitude and longitude in an Earthcentred Earthfixed (ECEF) system. Computationally, it can be expensive and slow to determine the total Earth radiation flux incident on a spacecraft, from the part of the Earth’s surface that is visible to that spacecraft, based on a search of the full CERES grid. To overcome this, we have developed a configurable Earth radiation model that reorganises the CERES data into a grid of triangles wrapped around the TOA surface. The number of triangles used to represent the TOA surface is configured during runtime, based on the number of triangles required to achieve a specified precision level. This approach is outlined in Li et al. (2017). At GNSS altitudes, radiation flux from the Earth is about 15 Wm^{−2}.

\( \varvec{F}_{{\mathbf{n}}} \) is the normal force acting in the direction of the surface normal, \( \hat{\varvec{n}} \),

\( \varvec{F}_{{{\mathbf{s}}^{\prime } }} \) is the shear force acting in the \( \hat{\varvec{s}}^{\prime } \) direction, which is along the projection of the total force onto the surface plane,

\( \varvec{F}_{{{\mathbf{mli}}}} \varvec{ } \) is the force due to the thermal reradiation from the MLI on the bus surface, which also acts along the normal direction,

E is the mean irradiance of the radiation source at one astronomical unit,

A is the area of the surface (determined in this case by the pixel array spacing),

c is the speed of light in vacuum,

\( \nu \) is the reflectivity of the material,

\( \mu \) is the specularity of the material,

\( \theta \) is the angle of incidence of the radiation with respect to the surface,

\( T_{\text{mli}} \) is the temperature of the MLI,

\( \sigma \) is the Stefan–Boltzmann constant,

\( \alpha \) is the absorptivity of the MLI material,

\( \epsilon_{\text{mli}} \) is the emissivity of the MLI material,

\( \epsilon_{\text{eff}} \) is the effective emissivity between the MLI and the spacecraft and

\( T_{\text{sc}} \) is the internal temperature of the spacecraft bus.
Equations 1 and 2 are also used for computing the SRP and ERP forces acting on the solar panels, but the spacecraft bus and the solar array are treated separately during force model computation, with the results combined during model implementation, as explained in Sect. 4. A similar approach is used by Darugna et al. (2018), and this is done because it simplifies the model computation process as it is not always practical to incorporate the correct solar panel behaviour into the raytracing computations.
2.1 Bus model computation scheme
2.2 Producing the grid files
The spiral points are not regularly spaced in latitude and longitude. Instead, the points are sorted according to distance along the spiral path, starting at the north pole (\( \varphi = 90^\circ , \lambda = 0^\circ \)) and ending at the south pole (\( \varphi =  90^\circ , \lambda = 0^\circ \)). This is not a standard method for organising the data. Thus, to provide a final model that is easily integrated into the POD processes of model users, we produce a set of acceleration grids, with grid nodes uniformly spaced at 1° intervals in latitude and longitude in the satellite frame.
 (i)
The radiation pressure model is computed using the spiral points scheme and the EPSsweep scheme.
 (ii)
The 10,000 spiral points data set is expanded using a padding process, see below.
 (iii)
Using modified Shepard’s method, 1600 grids are produced from the output of the spiral points computation, for each acceleration component, with all combinations of Q, W pairs considered, where both Q and W range from 11 to 50.
 (iv)For each grid file, the interpolated values at each of the 3960 EPSsweep points are calculated, and the interpolated value is compared with the results from the EPSsweep computation. This is used to compute the RMS error value, \( E_{\text{rms}} \), for that grid file according to:where \( a_{{{\text{EPS}},i}} \) are accelerations at point i according to the EPSsweep computation and \( a_{{{\text{grid}},i}} \) is an interpolated acceleration at point i derived from the grid file.$$ E_{\text{rms}} = \sqrt {\frac{1}{3960}\mathop \sum \limits_{k = 1}^{3960} (a_{{{\text{EPS}},i}}  a_{{{\text{grid}},i}} )^{2} } , $$(7)
 (v)
Finally, the grid files that minimise the \( E_{\text{rms}} \) quantity for each component (X, Y and Z) are chosen as the optimal grid files for that spacecraft.
2.3 Padding the spiral points data set
2.4 Modelling limitations
 (a)
Mismodelling of reflected radiation coming off the bus onto the panels and shadowing from the bus onto the panels (and vice versa). Both of these are due to the separate treatment of the solar panels and the spacecraft bus during model computation. Here, there is a tradeoff in modelling accuracy between being able to deal with nonstandard solar panel orientations and being able to capture the effects of reflections and selfshadowing of the bus onto the panels. An analysis of this tradeoff is not presented here but will be considered carefully in future development work.
 (b)
No modelling of the timeevolution of the surface material properties.
 (c)
Incomplete modelling of TRR effects. In the raytracing algorithm, we only consider spacecraft bus surfaces that are covered in multilayer insulation (MLI). This strategy can perform reasonably well on those satellites where the surfaces are mostly covered in MLI, as is the case with the GPS Block IIR and IIRM bus surfaces. However, it is limited in cases where a significant proportion of the spacecraft surface is not covered in MLI (e.g. the SAR antenna on Sentinel1; radiators on Galileo spacecraft, etc.). Also, we are not considering the force due to the temperature gradient across the Sunfacing and antiSunfacing sides of the solar panels in this study, but this effect has been considered in previous studies (Adhya 2005) and we are working towards developing a simplified approach to account for this.
 (d)
No modelling of thermal recoil forces due to emissions from radiators and other thermal control system components that actively emit heat. The impact of this will be different between the Block IIR and the Block IIRM satellites. The thermal control system of the Block IIRM satellites was updated with additional integral heat pipes due to high heat concentrations in the honeycomb structure of the Lband panel due, in part, to increased signal power needs (Hartman et al. 2000).
3 Model implementation
The UCL radiation force model implementation requires several inputs. Most of these are spacecraftspecific information that includes position, nominal mass, actual mass if available, the grid files for the bus model, solar panel properties (area, surface material properties), attitude information (in the form of attitude control laws or onboard attitude measurements) to enable accurate determination of the spacecraft BFS and solar panel orientation in the BFS.

\( m_{\text{n}} \) is the nominal mass of the spacecraft, i.e. the value used to compute the grid in kg,

\( m_{\text{a}} \) is the actual mass of the spacecraft in kg,

\( \varvec{\ddot{x}}_{\text{grid}} \) are the grid file accelerations in the spacecraft BFS x, y and zaxes in ms^{−2},

\( \widetilde{{\varvec{\ddot{x}}}}_{grid} \) are denormalised grid file accelerations in ms^{−2},

E is the mean solar irradiance at 1 AU.
4 The GPS IIR/IIRM model description and data sources
The computation of the force models for the bus is performed using Version 5.05 of UCL’s Analytical SRP and TRR Modelling Software at a nominal spacecraft mass of 1100 kg and a pixelarray resolution of 1 mm^{2}. The bus model grid files for the IIR/IIRM spacecraft, with and without the NAP antenna, are provided alongside this article as an electronic supplement.
GPS IIR/IIRM solar panel and yoke arm surface material properties that are used in the UCL radiation pressure models
Properties  Value 

Solar array  
\( A_{\text{panel}} \,\left( {{\text{m}}^{2} } \right) \)  13.59 
\( \nu_{\text{front}} \)  0.28 
\( \nu_{\text{rear}} \)  0.11 
\( \mu_{\text{front}} \)  0.85 
\( \mu_{\text{rear}} \)  0.50 
Yoke arms  
\( A_{\text{yoke}} \;\left( {{\text{m}}^{2} } \right) \)  0.32 
\( \nu_{\text{yoke}} \)  0.85 
\( \mu_{\text{yoke}} \)  0.85 
Statistics relating to the selection of the grid files that represent the radiation pressure model for the GPS IIR/IIRM spacecraft bus
Grid  Q  W  RMS error  Max error  Bias 

GPS IIR/IIRM, with NAP antenna  
X  37  13  0.0304  0.247  0.000370 
Y  50  50  0.0351  0.320  − 0.000610 
Z  32  11  0.1580  2.710  0.002210 
GPS IIR/IIRM, no NAP antenna  
X  34  13  0.0309  0.240  0.000353 
Y  50  50  0.0352  0.323  − 0.000858 
Z  32  11  0.1590  2.726  0.002202 
For the BW model, we use the values used by the European Space Operations Centre (ESOC) in their POD processing for the International GNSS Service (IGS; Johnston et al. 2017), which are 4.11, 0.0 and 4.25 m^{2} for the x, y and z faces of the bus, respectively (GarciaSerrano et al. 2016). The combined surface area of the solar array and yoke arms is 13.91 m^{2}. In the ESOC BW model, for the surface properties of the bus, \( \nu = 0.06 \) and \( \mu = 0 \). The solar panel properties of the BW model and the UCL model are identical.
For the antenna thrust model, we use the IGS model values for signal power (http://acc.igs.org/orbits/thrustpower.txt), which are 85 W for GPS Block IIR and 108 W and 198 W for GPS Block IIRM satellites.
5 Model validation
We investigate the performance of the new modelling strategy using two software systems: the UCL Orbit Dynamics Library (UCLODL) and ESOC’s Navigation Package for Earth Observation Satellites (NAPEOS) software (Springer 2009). The UCLODL comprises a set of programs developed by researchers at UCL over the years, for the explicit purpose of studying the impact of force modelling strategies that are developed by the UCL Space Geodesy and Navigation Laboratory. NAPEOS is a GNSS data processing package developed by ESOC and used in its contributions to IGS activities to produce satellite orbits, precise clocks, station coordinates, Earth rotation parameters and so on.
5.1 Analysis of the impact of separate model components using the UCLODL
Using the UCLODL, we performed a series of sensitivity analyses to investigate the impact of the individual model components and verify the implementation method. In these tests, as the reference trajectory, we used precise IGS final orbits, considering all available IIR and IIRM satellites over the full month of March 2016. For each satellite, we perform multiple orbit predictions, with separate prediction runs corresponding to separate IGS final orbit files. As such, in this part of the analysis, we consider 13 GPS IIR satellites and 7 GPS IIRM satellites. For those satellites with a complete set of IGS final orbits during the analysis period, we perform 31 prediction runs from 1 to 31 March 2016. In the orbit propagator, the general force modelling strategy uses Earth Gravity Model 2008 up to degree and order 20 (Pavlis et al. 2012, 2013) and the JPL Development Ephemerides 405 (DE405) for thirdbody gravitational forcing due to the Sun, Moon, Jupiter and Venus (Standish 1998). The solid Earth tide effect due to the Sun and the Moon is accounted for according to Marsh et al. (1987). General relativistic effects are modelled according to Sect. 3.7.3 of Montenbruck and Gill (2000). The numerical integration is based on an 8th order Runge–Kutta integrator.

Base model: SRPonly model using the ESOC BW model (GarciaSerrano et al. 2016).

Test 1: SRPonly, where the bus model comprises grids produced by the UCL raytracing software, but only Eqs. 1 and 2 are used.

Test 2: Same as Test 1, but here the bus model comprises grids that account for both SRP and the effects of MLI TRR (Eqs. 3 and 4).

Test 3: Same as Test 2, but with ERP turned on.

Test 4: Same as Test 3, but with AT turned on.
Orbit prediction error statistics for all GPS IIR and IIRM satellites on orbit in March 2016 (i.e. mean value across all satellites) with a range of radiation force modelling strategies considered
Modelling strategy  Orbit prediction error over 12h arc (m)  

RMS  Maximum 3d  
3d  Radial  Acrosstrack  Alongtrack  
No radiation force model  44.272  14.951  2.711  41.150  96.131 
Base model: SRPonly, BW  5.655  1.907  0.355  5.253  12.418 
Test 1: SRPonly, grids  2.990  1.020  0.188  2.769  6.685 
Test 2: SRP and Bus MLI TRR, grids  0.674  0.220  0.052  0.627  1.443 
Test 3: Test 2 + ERP  0.650  0.220  0.052  0.599  1.447 
Test 4: Test 3 + AT  0.648  0.220  0.052  0.598  1.440 
Orbit prediction error statistics for all GPS IIR and IIRM satellites on orbit in March 2016
PRN  SVN  Orbit prediction error over 12h arc  

RMS  Maximum  
3d  Radial  Acrosstrack  Alongtrack  3d  
IIR  
2  61  0.871  0.237  0.062  0.835  1.851 
4*  49  1.325  0.290  0.028  1.293  2.380 
11  46  0.632  0.219  0.064  0.588  1.586 
13  43  0.137  0.038  0.072  0.107  0.245 
14  41  0.923  0.244  0.070  0.888  1.921 
16*  56  0.563  0.219  0.027  0.516  1.426 
18*  54  0.731  0.220  0.019  0.687  1.367 
19  59  1.143  0.333  0.044  1.093  2.560 
20*  51  0.545  0.207  0.037  0.497  1.365 
21  45  0.273  0.126  0.034  0.237  0.685 
22*  47  0.432  0.175  0.030  0.392  1.125 
23  60  0.574  0.275  0.076  0.496  1.706 
28*  44  0.688  0.158  0.038  0.669  1.288 
IIRM  
5*  50  0.760  0.256  0.029  0.712  1.740 
7  48  0.689  0.171  0.088  0.662  1.133 
12*  58  0.471  0.234  0.059  0.404  1.024 
15  55  0.759  0.390  0.048  0.649  2.189 
17  53  0.239  0.160  0.068  0.164  0.786 
29  57  0.844  0.186  0.080  0.819  1.404 
31  52  0.371  0.260  0.079  0.252  1.023 
Mean  0.648  0.220  0.052  0.598  1.440  
SD  0.295  0.076  0.021  0.302  0.575 
5.2 Analysis of the impact of the new bus model on POD using NAPEOS
To assess the impact of introducing our gridbased model of the spacecraft bus on the quality of orbit estimates, we ran a number of POD analyses using NAPEOS. The analysis uses 100 tracking stations of the IGS MultiGNSS Experiment (MGEX) (Montenbruck et al. 2017b) and all observed GPS satellites, but the results presented here focus on the 13 GPS Block IIR and 7 Block IIRM satellites that were on orbit during the analysis period. The data processing method broadly follows ESOC’s IGS analysis strategy (ftp://igs.org/pub/center/analysis/esa.acn) where the basic observables are undifferenced carrier phases and pseudoranges and the integer carrier phase ambiguities are resolved (Ge et al. 2005). The Earth gravity model used is EIGENGL05C up to degree and order 12 (Foerste et al. 2008), and the JPL Development Ephemerides 405 (DE405) is used for thirdbody gravitational forcing due to the Sun, Moon and all solar system planets including Pluto (Standish 1998). The effects of solar Earth tides, ocean tides, solid Earth pole tide, oceanic pole tide and general relativistic corrections are accounted for according to the IERS conventions (Petit and Luzum 2010). The numerical integration uses the Adams–Bashforth/Adams–Moulton 8th order prediction–correction multistep method, as described in Springer (2009).
With the core data processing strategy fixed, we run the POD process using four different orbit modelling strategies, batch processed at 24h intervals, from 00:00:00 to 23:59:30 in GPS time, thus completely independent from day to day. For the orbits, we generate estimates for the midnight epoch, such that there is an overlap between consecutive solutions at a single point. A full year (2016) is considered, so there are 366 independent solutions and 365 overlap points. In addition to the orbit model parameters, station coordinates and Earth rotation parameters are also estimated. The orbit models considered include:
 1.
ECOM: No a priori radiation force model, only the reduced ECOM (Springer et al. 1999) and three constrained alongtrack parameters (constant, cosine and sine with argument of latitude as argument). Here, the alongtrack parameters are included as soakup parameters to absorb the effects of orbit mismodelling, which tends to manifest strongly in the alongtrack direction, as the results of Sect. 5.1 demonstrate.
 2.
ECOM + BW: Same estimation strategy as ECOMonly, but here we also include an a priori radiation force model using the ESOC BW model of the GPS IIR and IIRM spacecraft (GarciaSerrano et al. 2016).
 3.
ECOM + UCL: Here, the only difference with the ECOM + BW strategy is that the box is replaced by the gridbased model.
 4.
ECOM2: No a priori radiation force model, only the D4B1 extended ECOM (Arnold et al. 2015) along with the three constrained alongtrack parameters for consistency. Here, our analysis with the ECOM2 model is not as comprehensive as it might be (as it was not in the scope of our original study plan). This will be addressed in our future work.
Statistics of daily, independent estimates of the ECOM parameters from three orbit modelling strategies for GPS Block IIR and IIRM satellites over the full year of 2016
ECOM  ECOM + BW  ECOM + UCL  

D _{ 0}  
Mean  − 100.792  − 4.256  − 0.749 
RMS  0.548  0.165  0.161 
Y _{ 0}  
Mean  0.397  0.387  0.489 
RMS  0.134  0.124  0.082 
B _{ 0}  
Mean  0.603  0.572  0.006 
RMS  0.343  0.359  0.291 
Bsin  
Mean  − 1.260  − 0.188  0.430 
RMS  0.710  0.395  0.428 
Bcos  
Mean  0.281  0.281  0.284 
RMS  0.102  0.098  0.100 
A _{ 0}  
Mean  − 0.019  − 0.009  − 0.009 
RMS  0.104  0.088  0.062 
Asin  
Mean  − 0.009  − 0.007  − 0.006 
RMS  0.066  0.037  0.035 
Acos  
Mean  − 0.001  − 0.002  0.001 
RMS  0.073  0.041  0.038 
Orbit overlap difference statistics from a comparison of GPS Block IIR and IIRM orbits produced for daily, independent estimates using four different orbit modelling strategies over 2016
ECOM  ECOM + BW  ECOM + UCL  ECOM2  

Radial  
Mean  2.68  − 0.30  − 1.16  0.09 
RMS  24.19  21.10  18.82  20.71 
Alongtrack  
Mean  − 4.74  − 2.79  − 2.82  2.03 
RMS  32.00  27.94  26.37  26.79 
Crosstrack  
Mean  − 0.02  − 0.19  − 0.40  0.13 
RMS  23.28  19.47  18.17  18.41 
Error statistics for day 1 and day 14 orbit predictions for GPS Block IIR and IIRM orbits produced using three different orbit modelling strategies for 2016
1st day  14th day  

ECOM  ECOM + BW  ECOM + UCL  ECOM  ECOM + BW  ECOM + UCL  
Radial  
Mean  0.37  0.29  0.07  − 1.14  − 1.70  − 1.71 
RMS  2.38  1.64  1.85  93.64  65.16  44.81 
Alongtrack  
Mean  0.22  0.17  − 0.08  149.11  182.28  153.13 
RMS  7.68  6.95  4.75  1847.44  1819.45  1834.97 
Crosstrack  
Mean  − 0.10  0.00  0.00  − 0.20  − 0.02  − 0.01 
RMS  3.74  3.50  1.69  44.81  36.92  23.13 
Comparing RMS orbit prediction errors using ECOM + UCL against ECOM + BW, after 1day, we see the errors increase by 0.21 cm in the radial direction but fall by 2.20 cm and 1.81 cm in the alongtrack and crosstrack directions, respectively. For the 14th day predictions, we see a reduction in the RMS orbit prediction errors of 20.35 cm and 13.79 cm in the radial and crosstrack directions, but an increase of 15.52 cm in the alongtrack direction. Overall, these results suggest ECOM + UCL is outperforming ECOM + BW, in the day 1 and day 14 orbit prediction tests, but there are limitations to this analysis that should be addressed in future work for improved confidence in our findings. For example, because we use it as our reference model, it is possible that ECOM + UCL is favoured in these tests. Also, systematic errors, such as those that depend on the elevation of the Sun above the orbital plane, do not show up in the yearly statistics. A more complete picture of the comparative performance of the models should be investigated through time series analysis.
6 Conclusions and discussion
Recent developments to our radiation force modelling strategy were analysed using a new model for the GPS Block IIR and Block IIRM satellites. Advances to our approach include: an enhanced bus model computation scheme (based on the spiral points algorithm) that uses raytracing to determine the radiation fluxspacecraft interaction from 10,000 points distributed uniformly on a sphere surrounding the spacecraft; an improved method (from a numerical stability perspective) for producing grids spaced at \( 1^\circ \times 1^\circ \) intervals in latitude and longitude in the spacecraft frame using a padding process to extend the spiral points data in all directions to reduce the impact of edge effects; a quality assurance process that uses results from an EPSsweep computation with 3960 points for selecting an optimal set of grids. The models produced, and the proposed implementation method, were refined using a series of verification tests within the UCLODL. The impact of introducing UCL’s gridbased model into a full POD process was investigated by analysing the statistics of estimated orbit model parameters, orbit overlaps and orbit prediction errors. Combined, the results provide a good indication that introducing highfidelity analytical force modelling into the POD process can improve the quality of the estimated orbits and further refinements of the approach to address current limitations are worth pursuing.
One of the difficulties with the highfidelity approach lies in acquiring the spacecraft data (geometry, surface material properties, attitude history, mass and mass history) that is required to produce accurate models. It is hoped that the results in this paper adds evidence to the case for making this data available to the science and engineering community, where it is possible—especially the detailed geometry and surface material properties. Using an accurate spacecraft model, it is possible to compute the highfidelity radiation force model. However, the model computation time remains a problem and limits the number of development and testing cycles that we are able to perform. A typical model computation involves a \( 5 \times 5 {\text{m}}^{2} \) pixel array projected onto the spacecraft model at a 1 mm pixel spacing. In such a case, there are \( 2.5 \times 10^{7} \) rays per incoming radiation flux direction, \( 1 \times 10^{4} \) different directions in the spiral points computation scheme, and so this requires \( 2.5 \times 10^{11} \) rayspacecraft surface interaction calculations. As it stands, this process takes ~ 3 days to compute (job runtime as opposed to CPU time) on the UCL highperformance computing facility, Legion@UCL (can take longer when the facility is under heavy load), followed by ~ 1 day of analyst’s time to work through the process of generating the grids. Therefore, it is worth exploring methods for reducing model production times. We are beginning to explore the use of a graphical processing unit (GPU) to exploit standard computer graphics techniques in the computation of the radiation fluxspacecraft surface interaction—a process that naturally lends itself to being parallelised. This idea is demonstrated in Grey et al. (2017) where an OpenCL implementation of a radiation sourcesatellite surface interaction model that includes accurate modelling of diffuse reflection and apparent size of illumination source is used to simulate the impact of photoelectron emission on spacecraft surface charging. Also, we are exploring the use of an algorithm that reorganises the UCL spacecraft model components into a kdimensional tree data structure, to speed up the raytracing algorithm by greatly reducing the number of raysurface interaction tests that need to be performed (Li et al. 2018).
Notes
Acknowledgements
This work is supported financially by the UK Natural Environment Research Council (NE/K010816/1; Grant No. 157630). The collaboration between University College London and PosiTim to test new force modelling strategies in an orbit estimation process is funded, in part, through a European Space Agency project to develop force models for Galileo GNSS spacecraft. The authors acknowledge the use of the UCL Legion HighPerformance Computing Facility (Legion@UCL), and associated support services, in the completion of this work. The authors wish to thank three anonymous reviewers for their valuable comments and suggestions which have greatly improved the quality of this paper.
References
 Adhya S (2005) Thermal reradiation modelling for precise prediction and determination of spacecraft orbits. PhD Thesis, University College LondonGoogle Scholar
 Arnold D, Meindl M, Beutler G et al (2015) CODE’s new solar radiation pressure model for GNSS orbit determination. J Geod 89:775–791. https://doi.org/10.1007/s0019001508144 CrossRefGoogle Scholar
 BarSever Y, Kuang D (2003) New empiricallyderived solar radiation pressure model for GPS satellites. JPL TRS 1992+Google Scholar
 Beutler G, Brockmann E, Gurtner W et al (1994) Extended orbit modeling techniques at the CODE processing center of the international GPS service for geodynamics (IGS): theory and initial results. Manuscr Geod 19:367–386Google Scholar
 Cerri L, Berthias JP, Bertiger WI et al (2010) Precision orbit determination standards for the Jason Series of Altimeter Missions. Mar Geod 33:379–418CrossRefGoogle Scholar
 Colombo OL (1986) Ephemeris errors of GPS satellites. Bull Géod 60(1):64–84. https://doi.org/10.1007/BF02519355 CrossRefGoogle Scholar
 Darugna F, Steigenberger P, Montenbruck O, Casotto S (2018) Raytracing solar radiation pressure modeling for QZS1. Adv Sp Res 62(4):935–943. https://doi.org/10.1016/J.ASR.2018.05.036 CrossRefGoogle Scholar
 Doornbos E, Scharroo R, Klinkrad H et al (2002) Improved modelling of surface forces in the orbit determination of ERS and ENVISAT. Can J Remote Sens 28(4):535–543. https://doi.org/10.5589/m02055 CrossRefGoogle Scholar
 Fliegel HF, Gallini TE (1996) Solar force modeling of block IIR Global Positioning System satellites. J Spacecr Rockets 33(6):863–866. https://doi.org/10.2514/3.26851 CrossRefGoogle Scholar
 Foerste CU, Flechtner F, Schmidt R et al (2008) EIGENGL05C—a new global combined highresolution GRACEbased gravity field model of the GFZGRGS cooperation. In: Geophysical research abstracts. Vienna, Austria, vol 10, Abstract No EGU2008A06944Google Scholar
 Franke R, Nielson G (1980) Smooth interpolation of large sets of scattered data. Int J Numer Methods Eng 15(11):1691–1704CrossRefGoogle Scholar
 GarciaSerrano C, Springer T, Dilssner F et al (2016) ESOC’s multiGNSS processing. In: IGS workshop 2016. Sydney, AustraliaGoogle Scholar
 Ge M, Gendt G, Dick G, Zhang FP (2005) Improving carrierphase ambiguity resolution in global GPS network solutions. J Geod 79:103–110. https://doi.org/10.1007/s0019000504470 CrossRefGoogle Scholar
 Gini F (2014) GOCE precise nongravitational force modeling for POD applications. Ph.D. Thesis, University of PaduaGoogle Scholar
 Grey S, Ziebart M (2014) Developments in high fidelity surface force models and their relative effects on orbit prediction. In: AIAA/AAS astrodynamics specialist conference. American Institute of Aeronautics and AstronauticsGoogle Scholar
 Grey S, Marchand R, Ziebart M, Omar R (2017) Sunlight illumination models for spacecraft surface charging. IEEE Trans Plasma Sci 45(8):1898–1905. https://doi.org/10.1109/TPS.2017.2703984 CrossRefGoogle Scholar
 Hartman T, Boyd LR, Koster D et al (2000) Modernizing the GPS Block IIR Spacecraft. In: ION GPS 2000, pp 2115–2121Google Scholar
 Hastings D, Garrett H (1996) Spacecraft–environment interactions. Cambridge University Press, CambridgeCrossRefGoogle Scholar
 Johnston G, Riddell A, Hausler G (2017) The International GNSS Service. In: Teunissen PJG, Montenbruck O (eds) Springer Handbook of Global Navigation Satellite Systems, 1st edn. Springer International Publishing, Cham, pp 967–982. https://doi.org/10.1007/9783319429281_33 CrossRefGoogle Scholar
 Klinkrad H, Koeck C, Renard P (1991) Key features of a satellite skin force modelling technique by means of MonteCarlo ray tracing. Adv Sp Res 11(6):147–150. https://doi.org/10.1016/02731177(91)90244E CrossRefGoogle Scholar
 Li Z, Ziebart M, Grey S, Bhattarai S (2017) Earth radiation pressure modelling for BeiDou IGSO Satellites. In: China Satellite Navigation conference, Shanghai, 2017Google Scholar
 Li Z, Ziebart M, Bhattarai S et al (2018) Fast solar radiation pressure modelling with ray tracing and multiple reflections. Adv Sp Res 61(9):2352–2365. https://doi.org/10.1016/j.asr.2018.02.019 CrossRefGoogle Scholar
 Marsh G, Lerch FJ, Putney BH et al (1987) an improved gravitational model of the Earth’s field: GEMTI. NASATM4019, REPT87B0451, NAS 1.15:4019, NASA, USAGoogle Scholar
 Marshall JA, Luthcke SB (1994) Modeling radiation forces acting on Topex/Poseidon for precision orbit determination. J Spacecr Rockets 31(1):99–105. https://doi.org/10.2514/3.26408 CrossRefGoogle Scholar
 Meindl M, Beutler G, Thaller D et al (2013) Geocenter coordinates estimated from GNSS data as viewed by perturbation theory. Adv Sp Res 51(7):1047–1064. https://doi.org/10.1016/J.ASR.2012.10.026 CrossRefGoogle Scholar
 Montenbruck O, Gill E (2000) Satellite orbits : models, methods and applications. Springer, HeidelbergCrossRefGoogle Scholar
 Montenbruck O, Steigenberger P, Hugentobler U (2015) Enhanced solar radiation pressure modeling for Galileo satellites. J Geod 89:283–297. https://doi.org/10.1007/s0019001407740 CrossRefGoogle Scholar
 Montenbruck O, Steigenberger P, Darugna F (2017a) Semianalytical solar radiation pressure modeling for QZS1 orbitnormal and yawsteering attitude. Adv Sp Res 59(8):2088–2100. https://doi.org/10.1016/J.ASR.2017.01.036 CrossRefGoogle Scholar
 Montenbruck O, Steigenberger P, Prange L et al (2017b) The MultiGNSS Experiment (MGEX) of the International GNSS Service (IGS)—achievements, prospects and challenges. Adv Sp Res 59(7):1671–1697. https://doi.org/10.1016/J.ASR.2017.01.011 CrossRefGoogle Scholar
 Pavlis NK, Holmes SA, Kenyon SC, Factor JK (2012) The development and evaluation of the Earth Gravitational Model 2008 (EGM2008). J Geophys Res Solid Earth 117:B04406. https://doi.org/10.1029/2011JB008916 CrossRefGoogle Scholar
 Pavlis NK, Holmes SA, Kenyon SC, Factor JK (2013) Erratum: correction to the development and evaluation of the earth gravitational model 2008 (EGM2008). J Geophys Res Solid Earth 118(5):2633. https://doi.org/10.1002/jgrb.50167 CrossRefGoogle Scholar
 Petit G, Luzum B (eds) (2010) IERS Conventions (2010) (IERS Technical Note 36). Verlag des Bundesamts für Kartographie und Geodäsie, Frankfurt am MainGoogle Scholar
 Renka RJ (1988) Multivariate interpolation of large sets of scattered data. ACM Trans Math Sci 4(2):139–148CrossRefGoogle Scholar
 RodríguezSolano CJ (2009) Impact of Albedo Modelling on GPS Orbits. Master Thesis, Technical University of MunichGoogle Scholar
 RodriguezSolano CJ, Hugentobler U, Steigenberger P (2012) Adjustable boxwing model for solar radiation pressure impacting GPS satellites. Adv Sp Res 49(7):1113–1128. https://doi.org/10.1016/J.ASR.2012.01.016 CrossRefGoogle Scholar
 Saff EB, Kuijlaars ABJ (1997) Distributing many points on a sphere. Math Intell 19(1):5–11CrossRefGoogle Scholar
 Shepard D (1968) A twodimensional interpolation function for irregularlyspaced data. In: Proceedings1968 ACM national conferenceGoogle Scholar
 Sibthorpe AJ (2006) Precision nonconservative force modelling for low earth orbiting spacecraft. Ph.D. Thesis, University College LondonGoogle Scholar
 Sośnica K, Thaller D, Dach R et al (2015) Satellite laser ranging to GPS and GLONASS. J Geod 89(7):725–743. https://doi.org/10.1007/s0019001508108 CrossRefGoogle Scholar
 Springer T (2009) NAPEOS mathematical models and algorithms. DOPSSYSTN0100OPSGN, Issue 1.0, Nov 2009. http://navigationoffice.esa.int/products/napeos
 Springer TA, Beutler G, Rothacher M (1999) A new solar radiation pressure model for GPS satellites. GPS Solut 2(3):50–62. https://doi.org/10.1007/PL00012757 CrossRefGoogle Scholar
 Standish EM (1998) JPL planetary and lunar ephemerides: DE405/LE405. Interoffice Memorandum 312.F98048, Jet Propulsion LaboratoryGoogle Scholar
 Tan B, Yuan Y, Zhan B, et al (2016) A new analytical solar radiation pressure model for current BeiDou satellites: IGGBSPM. Sci Rep 6(32967): https://doi.org/10.1038/srep32967
 Tapley BD, Schutz BE, Born GH (2004) Statistical orbit determination. Academic Press, CambridgeGoogle Scholar
 Vallado DA, McClain WD (2001) Fundamentals of astrodynamics and applications. Kluwer Academic Publishers, DordrechtGoogle Scholar
 Wang C, Guo J, Zhao Q, Liu J (2018) Solar radiation pressure models for BeiDou3 I2S satellite: comparison and augmentation. Remote Sens 10(1):118. https://doi.org/10.3390/rs10010118 CrossRefGoogle Scholar
 Wielicki BA, Barkstrom BR, Harrison EF et al (1996) Clouds and the Earth’s Radiant Energy System (CERES): an Earth observing system experiment. Bull Am Meteorol Soc 77(5):853–868. https://doi.org/10.1175/15200477(1996)077%3c0853:CATERE%3e2.0.CO;2 CrossRefGoogle Scholar
 Willis P, Beutler G, Gurtner W et al (1999) IGEX: international GLONASS experiment—scientific objectives and preparation. Adv Sp Res 23(4):659–663. https://doi.org/10.1016/S02731177(99)001477 CrossRefGoogle Scholar
 Zelensky NP, Lemoine FG, Ziebart M et al (2010) DORIS/SLR POD modeling improvements for Jason1 and Jason2. Adv Sp Res 46(12):1541–1558. https://doi.org/10.1016/J.ASR.2010.05.008 CrossRefGoogle Scholar
 Ziebart M (2001) High precision analytical solar radiation pressure modelling for GNSS spacecraft. Ph.D. Thesis, University of East LondonGoogle Scholar
 Ziebart M (2004) Generalized analytical solar radiation pressure modeling algorithm for spacecraft of complex shape. J Spacecr Rockets 41(5):840–848. https://doi.org/10.2514/1.13097 CrossRefGoogle Scholar
 Ziebart M, Dare P (2001) Analytical solar radiation pressure modelling for GLONASS using a pixel array. J Geod 75:587–599. https://doi.org/10.1007/s001900000136 CrossRefGoogle Scholar
 Ziebart M, Adhya S, Sibthorpe A, Cross P (2003) GPS Block IIR nonconservative force modelling: computation and implications. In: ION GPS/GNSS 2003. Portland, OR, pp 2671–2678Google Scholar
 Ziebart M, Adhya S, Sibthorpe A et al (2005) Combined radiation pressure and thermal modelling of complex satellites: algorithms and onorbit tests. Adv Sp Res 36(3):424–430. https://doi.org/10.1016/j.asr.2005.01.014 CrossRefGoogle Scholar
 Ziebart M, Sibthorpe A, Cross P, et al (2007) Cracking the GPSSLR orbit anomaly. In: ION GNSS 2007. Fort Worth, Texas, pp 2033–2038Google Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.