Research of PALSAR DEM Denoising Based on Adaptive Gaussian Convolution

[Purposes] In order to solve the problem of data noise caused by DEM elevation distortion in mountain area, terrain factors were introduced into DEM data processing algorithms, and a terrain adaptive Gaussian convolution DEM denoising method was proposed. [Methods] In this research, four areas are discussed along with slope directions according to DEM elevation gradient direction, and the Gaussian second-order function formula was rotated. In areas with different slope orientations, by combining traversing 3-13 size windows and 0.5-10 standard deviation σx, σy with Python, and verifing by ICESat-2/ATL08 data, the best convolution window and standard deviation value of PALSAR DEM each region are determined according to the root mean square error. Finally, the results of noise reduction are evaluated. [Findings] The results show that the optimal convolution window size of PALSAR DEM is 9; the σ varies slightly in all directions; generally, σx does not exceed 6 and σy does not exceed 3.5. The root mean square error of DEM decreased by 7.00%, 11.49%, 9.24%, 8.55% in four directions severally. [Conclusions] The contour extracted from Gaussian denoised DEM are smoother than that before denoising, and the extracted hydrological network and watershed are more complete and reasonable.