The terrestrial ecosystem carbon monitoring satellite (CM-1) is China’s first forestry-focused remote sensing satellite designed to measure the vertical structure of terrestrial ecosystem forests. Equipped with a 5-beam LiDAR, the satellite collects high-precision ground elevation data. However, laser altimetry data is affected by atmospheric conditions and complex terrain during transmission, making it unsuitable for direct use as elevation control points in its raw form. To address this challenge, we develop an automatic classification extraction method for laser elevation control points tailored to CM-1. This method leverages the characteristics of satellite data and employs multi-criteria constraints to ensure high elevation accuracy, providing critical support for generating regional digital surface model (DSM) using stereo images.

To ensure the extracted laser elevation control points meet accuracy requirements, we propose a multi-criteria constraint-based automatic classification method. First, the elevation difference between the laser point and an open digital elevation model (DEM) is calculated, and laser points with differences exceeding 30 m are flagged as gross errors. Coarse screening is then conducted to assess data validity. The maximum amplitude of the echo waveform is analyzed to identify and eliminate saturated data, while low signal-to-noise ratio (SNR) data is also eliminated. Subsequently, only laser data with a single waveform peak is retained to mitigate the influence of complex ground surfaces on elevation accuracy. Using the laser radar equation, the relationship between surface slope, received waveform pulse width, and elevation accuracy is analyzed. The pulse width is employed to estimate the elevation accuracy of laser points, which are then classified based on different accuracy levels.

High-precision airborne laser point cloud data from Shenyang and Pennsylvania are used to validate the accuracy of extracted laser elevation control points. In Shenyang, 1353 laser points are initially identified, of which 778 are retained after screening. The overall elevation accuracy improves from 2.410 m to 0.440 m (Table 3). In Pennsylvania, 23713 laser points are identified with 5226 retained, resulting in an accuracy improvement from 4.130 m to 0.747 m (Table 4). The influence of different screening parameters, including DEM elevation difference, saturation, SNR, and waveform peak count, is statistically analyzed in the two test areas (Table 5). A regional DSM test is conducted by integrating laser elevation control points with stereo images. The results demonstrate a significant improvement in DSM accuracy, with elevation errors reduced from 11.45 m to 2.27 m.

In this paper, we first analyze the quality of multi-beam laser data from the CM-1. Based on the characteristics of the laser data, a multi-criteria constraint method for extracting laser elevation control points is developed, enabling classification by elevation accuracy. Validation in Shenyang and Pennsylvania demonstrates significant improvements in elevation accuracy. The errors of the extracted laser elevation control points are reduced from (0.099±2.410) m and (0.945±4.130) m to (-0.007±0.440) m and (-0.086±0.607) m, respectively. The extracted points meet elevation control requirements for 1∶50000 or larger scale stereo mapping. Moreover, integrating laser elevation control points with multi-angle block adjustment reduces the root mean square error (RMSE) of 10 m grid DSM, generated from ±19° images, from 11.45 m to 2.27 m, meeting the elevation accuracy requirements for 1∶50000 scale topographic mapping.