Enhancing marine gravity anomaly recovery in the South China Sea using back propagation neural network

SUMMARY Satellite altimetry technology can recover the marine gravity field by measuring Sea Surface Heights (SSHs) with high precision. However, traditional methods, relying on linearized SSH–gravity relations, fail to capture complex nonlinear characteristics. Meanwhile, the accuracy of the altimeter-only gravity field is often compromised in shallow depth areas due to poor-quality altimetric signals. To address these issues, this study proposes a method for recovering gravity anomalies that combines Back Propagation Neural Network (BPNN) with multisource data. The BPNN establishes a nonlinear relationship between gravity and input parameters, including Deflections of the Vertical (DOVs), Seafloor Topography (ST), Vertical Gravity Gradients (VGGs) and Gravity Anomalies (GAs), thereby constructing a gravity anomaly model for the South China Sea. For benchmarking, gravity is also derived with the classical Inverse Vening–Meinesz (IVM) method and validated against independent shipborne gravity which is applied to evaluate the performance of the gravity model. The results demonstrate that the BPNN method outperforms the IVM method, achieving an accuracy improvement of 1.08 mGal overall, and 1.61 mGal in shallow depth areas. Additionally, compared with the reference gravity models (SWOT and DTU17), the gravity model derived by the BPNN method achieves an accuracy improvement of 0.04 and 0.85 mGal, respectively. Power spectra analysis further reveals that the improvements from the BPNN method are most significant in the wavelength range of 5–100 km. The improved accuracy is attributed to the effective incorporation of ST, VGG and prior GA information. The results show that the BPNN method effectively captures nonlinear features and has significant potential for marine gravity field recovery.