Research on a multi-level adaptive denoising algorithm based on ICESat-2 background photon characteristics.

ICESat-2 data play a critical role in Earth observation, but the high sensitivity of the advanced topographic laser altimeter system (ATLAS) makes observations susceptible to environmental factors like solar radiation, generating substantial noise that compromises accurate retrieval of surface structural parameters. This paper proposes a preliminary-to-fine Multi-level denoising algorithm, whose innovations lie in improving the elliptical search domain in the fine denoising phase to adapt to slope change under complex terrain, innovatively proposing a method based on maximum background rate Indirect estimation threshold to eliminate noise, and in addition, optimizing the Box plot filter to make it more conforming to the photon distribution characteristics of areas with complex vertical structures such as forest and city. The outcome shows that the background rate of daytime ICESat-2 data is on the order of 10⁵-10⁶ points per second, which can directly reflect the density of noise photons in the data; in the data validation of four types of surface features, namely forest, wasteland, glacier, and city, the average denoising rate of the algorithm in this paper reaches 70%. Compared with existing algorithms, the R² is increased by at least 4.30%, the MAE is reduced by 0.68 m, and the RMSE is reduced by 0.95 m, which fully attests to the suitability of the algorithm for denoising ICESat-2 data in daytime strong noise environments.