The Ice

Cloud

and land Elevation Satellite-2（ ICESat-2） will use a multiple-beam photon-counting Li DAR system to observe the earth surface with profile model at 532 nm. Given that the photon point cloud with low point density captured at a 490 km high altitude is highly affected by noise

especially those from solar background

traditional filtering algorithms for point cloud are not appropriate to the profile photon returns.We propose an automatic filtering algorithm for the photon profile data acquired by the Multiple Altimeter Beam Experimental Li DAR（ MABEL） system

a high-altitude airborne profile-laser altimeter designed as a simulator for ICESat-2. In addition

the basic principle of the multiple-beam photon-counting system

the property of the photon point cloud

and the tree height retrieval method are discussed.The data processing of MABEL points in this paper mainly consists of three steps.（ 1） Denoising was implemented in the acquired MABEL data. The frequency histogram for the accumulated distance between the given point and its k-nearest neighbors was assumed as Gaussian-like distribution

whereas the point with its accumulated distance far from the mean distance was recognized as noise.（ 2） The designed filtering algorithm was applied on the denoised MABEL data. The points were divided into sections along the profile; the lowest point in each section was selected as initial ground returns. Ground points in a window that covers several sections were adopted for modeling the second order local terrain. Least square fitting method was used to determine the local curve parameters. The points in each section were classified into ground points and no-ground points using the adaptive threshold.（ 3） The labeled ground points were used to produce line-like Digital Elevation Model（ DEM） by an interpolation method. The mean vegetation height in the study area can be estimated by comparing the local maximal laser points with the line-like DEM.The point clouds acquired by the MABEL 532 nm channel at Sierras-Forest in September 2012 were tested with the filtering algorithm. Approximately 2001 points were acquired in the experiment area with 2310 m long. After applying the proposed filtering algorithm

760 points were labeled as ground

1159 points were labeled as vegetation

and the remaining 82 points were labeled as noise. Validation results showed the total classification accuracy of the proposed algorithm was 97. 6% compared with the manual method. In addition

the line-like DEM and the mean tree height（ 34. 2 m） in the study area were estimated; error analysis was also performed.This paper introduced the basic principles of multiple-beam photon-counting Li DAR of ICESat-2

a novel point cloud filtering algorithm for MABEL data to separate ground and vegetation points. Both the DEM and the mean tree height were estimated from the filtered data. The following conclusions were obtained:（ 1） The experiment shows that the filtering algorithm is efficient and self-adapting for separating the ground and vegetation MABEL point.（ 2） The noise contained in the MABEL data lead to failed labeling

and directly increases the errors of the estimated DEM and tree height.（ 3） Given the noise level of the ICESat-2 data may be higher than the MABEL data

further studies are needed on both data denosing and ground points detection.