Measuring and monitoring forest canopy height are essential for effective forest management and ecosystem assessment. We evaluated the accuracy of forest canopy height based on Digital Surface Models (DSMs) derived from three different optical earth observation satellite stereo images: WorldView-3 along-track stereo pair images, BlackSky Gen-2 along-track 5-shot stereo images, and Newsat cross-track multiple stereo images. BlackSky Gen-2 5-shot stereo images achieved the highest accuracy on the forest canopy height measurement. Over coniferous forest areas with minimal seasonal height variation, the accuracy metrics were as follows: for WorldView-3, the standard deviation was 1.87 meters, Root Mean Square Error (RMSE) was 2.12 meters, and Normalized Median Absolute Difference (NMAD) was 0.97 meters; for BlackSky Gen-2, the standard deviation was 1.65 meters, RMSE was 1.79 meters, and NMAD was 1.24 meters; and for Newsat, the standard deviation was 1.59 meters, RMSE was 1.84 meters, and NMAD was 1.34 meters. The topographic azimuth affected the decreasing accuracy of the DSM derived from the stereo pair images of WorldView-3 on slopes facing away from the satellite's azimuth. The results highlighted the importance of acquiring stereo images from multiple suitable viewing angles to ensure accurate forest measurements in mountainous area. These findings demonstrate the applicability of DSMs derived from satellite stereo images, particularly multi-view stereo images from small satellites, for measuring forest canopy height.
