Global Gravity Fields from Simulated Level-1 GRACE Data 147
no improvements in the coefficients could be achieved. If the remaining differences
after the gravity field adjustment stay between the baseline accuracy and the errors
introduced by the a priori field, the error sources studied in the experiment do play
a role in the real data processing and significantly add to the error budget of the
estimated gravity fields.
To assess the influence of the surface forces on the satellite trajectories, and
especially the effect of errors in these models on the estimation of gravity field coef-
ficients, the albedo model was set to zero in a next parameter estimation run. The
effect on the adjusted gravity field coefficients is dramatic. As shown in Fig. 1b,
the estimated parameters do not longer converge to the correct values. The dif-
ferences between adjusted and “true” gravity field model grow bigger than the
differences between the a priori and the “true” model. This emphasizes the neces-
sity of either good non-gravitational models, sufficient empirical accelerations to
compensate model deficiencies, or the direct measurement of the surface forces by
onboard accelerometers as in case of GRACE. No further experiments with direct
solar pressure or atmospheric drag were made, since the size of these surface forces
is even bigger than that of albedo.
In order to simulate the direct measurement of the surface forces by the onboard
accelerometers, the modeled forces were added up and stored as GRACE Level-1B
ACC1B observation data files. In a similar way, the simulated K-Band observa-
tions were written t o KBR1B files, and the attitude angles were transformed into
quaternions and stored as SCA1B files. The change to GRACE Level-1B format
includes a transformation between inertial coordinates and satellite system coordi-
nates. Moreover, some digits are lost due to the formatted output of the EPOS-OC
files. To check this effect, the parameter estimation run with GGM02C as a pri-
ori field was repeated, this time based on the simulated Level-1B observation files
instead of force models. The result of the gravity field recovery using GPS and
K-Band range-rate data is also shown in Fig. 1b. It does not differ significantly from
the results shown in Fig. 1a. The oscillations between even and odd low degree
coefficients are a phenomenon often observed in simulation studies in the absence
of observation noise and are commonly attributed to the influence of the polar gap
of the satellite orbits (GRACE has an inclination of 89.5
◦
). As a result of the tests
carried out so far, we conclude that the EPOS-OC software is able to simulate obser-
vations and to estimate gravity field coefficients with sufficient accuracy both for this
study and for the routine processing of GRACE data.
4 Estimation of Instrument Parameters
In the routine processing of real GRACE data some empirical parameters, referred
to as instrument parameters, are introduced to account for systematic effects like
drifts or periodic errors in the observation noise. In this section their correlation
with gravity field coefficients is described. The instrument parameters include scale
factors for the along-track accelerations (FAT) and accelerometer biases in along-
track (P1T), cross-track (P1N), and radial (P1R) direction that are estimated at the