Land elevation data are indispensable for topographic mapping and geological disaster monitoring. However, the
existing ICESat-2/ATL08 (V04) product has a coarse resolution (≥100 m) and is characterized by high uncer-
tainty in mountainous areas; thus, it cannot be used to describe terrain relief characteristics accurately. In this
study, a new method for extracting terrain surface elevation is proposed, which uses a local statistical denoising
algorithm for mountainous areas (LSDAMA) based on the raw georeferenced photon product ICESat-2/ATL03.
Coarse denoising is based on performing histogram thresholding, and refined denoising is based on local slope
fitting; the process of performing coarse denoising twice and then refined denoising not only improves the
removal effect for noise photons but also increases the signal photon retention rate in mountainous areas.
Additionally, the estimation accuracy of the terrain surface elevation can be improved by setting rectangular
dynamic windows along the fitted slope direction. Using the Babao River Basin in the Qilian Mountains as the
research area, a total of 137 validation points from 777 GPS CORS in 40 quadrats and UAV LiDAR measurements
were used to verify the accuracy. The results showed that the terrain surface elevations estimated by the
LSDAMA are more accurate than those estimated by the ATL08 official products, especially in mountainous areas
with slopes greater than 20◦. The root mean square error (RMSE) of the LSDAMA decreased from 2.72 m for the
ATL08 product to 0.60 m, and the mean deviation (MBE) decreased from − 1.27 to 0.04 m. Additionally, the
LSDAMA greatly improved the signal photon retention rate and reduced the interval between adjacent elevation
points from 100 m for the ATL08 product to 2–7 m; this reduced interval can be used to describe the terrain
fluctuation characteristics in detail, thus providing reliable basic data for monitoring terrain surface elevation
changes in mountainous areas.

Photon counting system