A machine learning approach to recover GRACE-B accelerometer data

In gravimetry satellite missions GRACE (Gravity Recovery and Climate Experiment) and GRACE-FO (GRACE Follow-On), accelerometer measurements from both satellites are necessary for the gravity field recovery. The accelerometer provides accurate measurements of the non-gravitational forces acting on the spacecraft, such as atmospheric drag, solar radiation pressure and albedo. These measurements are required to separate any non-gravitational effect from the sought-after gravitational perturbations on the spacecraft motion. Therefore, the quality of accelerometer data, denoted as ACC products, significantly affects the quality of gravity field models.

Near the end of the GRACE mission, due to the reduced battery capacity, the on-board accelerometer of the GRACE-B was turned off and its measurements were replaced by synthetic accelerometer data, called transplant data. The transplant data are generated by a series of adjustments to the GRACE-A ACC data. A similar approach was also employed for the GRACE-FO mission, when the GRACE-D ACC data degraded and were required to be replaced with synthetic data as well. Using the transplant data in both missions is one of the main challenges of providing high-quality gravity field models.

We investigate the feasibility of Machine Learning (ML) algorithms for the recovery of GRACE-B ACC based on GRACE-A measurements and orbital data such as shadow factor and β angle. Taking advantage of ~14 years of GRACE-B measurements, this work aims to develop a model which can predict the missing accelerometer data under different orbital conditions. Two different architectures are implemented to forecast GRACE-B accelerometer data: Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM). The performance is evaluated using the Root Mean Square Error (RMSE) and by comparing the predicted data with the calibrated real data in the evaluated period. Furthermore, the ML-based ACC products will be compared to the transplant products and their impact on the gravity field will be discussed.