Numerical experiment on downward continuation of airborne gravity in the Qinghai-Xizang Plateau

The Downward continuation is a critical step in the processing of airborne gravity data. Nevertheless, it tends to amplify errors and makes it difficult to obtain stable and reliable solutions. Here, we proposed three optimization methods to improve the least-squares (LS) solution, as modified Poisson’s kernel, regularization, and singular value decomposition (SVD), and the results are analysed contrastively. In addition, the spectral decomposition form of the generalized cross-validation (GCV) criterion for determining the regularization coefficient was deduced to improve computational efficiency. In this paper, we take the Qinghai-Xizang Plateau as the study area. The results show that the least-squares method can obtain a reliable continuation solution without taking error into account. However, with the addition of random noise, the least-squares continuation results are unusable for the severe ill-posedness of the normal equation. The standard deviation (STD) decreased from 97.888 mGal to 7.006 mGal after regularization, and the stability of the solutions has been significantly improved. In this case, after removing the low-order coefficient model and modifying Poisson’s kernel function, the accuracy of the continuation results is improved by 18%, and the far-zone effect is also reduced significantly, which can be ignored. The least squares solution based on SVD has obvious advantages in computational efficiency for selecting the regularization parameter. Compared with traditional inversion calculations, the calculation speed can be increased by 16 times.