Accuracy assessment of DSM in mountainous forest areas generated from GF-7 stereo images assisted by ATLAS/ICESat-2

Abstract. The successful launch of the GaoFen-7 (GF-7) satellite has made it possible to achieve sub-meter-level three-dimensional mapping. However, most current studies focus on accuracy evaluating of GF-7 stereo mapping in bare land, and less on its performance in mountainous forest areas. This study aims to explore the potential of ground control points (GCPs) from the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) to enhance the accuracy of GF-7 digital surface models (DSM) in forested areas, and further evaluate the performance of GF-7 stereo mapping in mountainous forest areas. First, the GCPs was extracted from ATL03 based on the direction adaptive OPTICS ground surface detection method. Then, with the assistance of Google Earth (GE), fine screening was applied to extract bare GCPs in forested areas, which were called ATLGE GCPs. Finally, the ATLGE GCPs were used as elevation control points, combined with GF-7 stereo images to generate the forest area DSM, which was called GF-7 DSM_ATLGE. To compare and analyze the potential of ATLGE GCPs in improving the accuracy of GF-7 forest DSM, SRTM and SLA03 were employed as elevation controls data to generate GF-7 DSM_SRTM and GF-7 DSM_SLA03, respectively. The elevation accuracy of the three DSMs was quantitatively evaluated using the DSM obtained from field-surveyed plots. The results showed that while GF-7 DSM_SRTM, GF-7 DSM_SLA03, and GF-7 DSM_ATLGE all effectively represented forest features, their accuracy varied significantly. Specifically, the elevation RMSE of GF-7 DSM_SRTM was 11.87 m, GF-7 DSM_SLA03 reached 4.78 m, and GF-7 DSM_ATLGE reached 2.99 m, showing a significant improvement. These results not only demonstrate the effectiveness of ATLGE GCPs in enhancing the elevation accuracy of GF-7 DSMs in forest areas, but also highlight the stereo mapping capability of GF-7 stereo images under controlled conditions in complex mountainous forest areas.