面向森林高度提取的光学多角度立体观测影像模拟

Canopy height is one of important indicators of carbon storage in forest ecosystems. Previous studies have demonstrated that optical stereo imagery held great potential for deriving forest canopy height. However, optical images are easily affected by cloud coverage and rains. The regional mapping of forest canopy height has to be achieved by the synthesis of multi-sensor and multi-temporal imagery. The spatial resolutions and viewing angles of existing spaceborne stereoscopic systems are quite different. It is essential to make a systematic investigation about the impact of image spatial resolutions and viewing angles on the vertical distribution of stereoscopic point clouds, which is the basis for the synergy of images acquired at different seasons even by different cameras. The theoretical model for the simulation of optical stereo imagery over forested areas is necessary for such studies. The LandStereo is such kind of model, which can simulate the along-track(only In-track viewing) stereoscopic imagery like ALOS/PRISM and ZY-3. However, the current version of LandStereo has no mode to simulate the along-track(including In-track and Cross-track) stereoscopic imagery like Worldview-1/2/3. Therefore, this study reported the modification of the LandStereo model to have it being able to simulate images acquired by any possible viewing directions in forest areas. Firstly, the calculation method of linear array projection center coordinates is improved, from originally only considering the change of observation altitude angle to considering the change of azimuth angle and altitude angle. Secondly, Rigorous imaging geometric model is improved to obtain RPC of images acquired by any possible viewing directions. Based on the improved LandStereo model, the bare and mountainous forest images with altitude angle of 75° and azimuth angle of 0°, 90°, and 225° are simulated to verify the accuracy and extract forest canopy height. The surface elevation extracted by the improved LandStereo model is consistent with the input DTM with high accuracy (r=0.99, RMSE=1.507), which proves the geometric accuracy of the improved LandStereo model. There are certain differences in the extraction accuracy of forest canopy height between stereo images of different angles. The results showed that the modified model could correctly simulate the stereoscopic features of forest canopy with given view direction, also initially demonstrated that view angle was an important factor affecting the estimation accuracy of forest canopy height by stereoscopic images. © 2023 Science Press. All rights reserved.

Simulation of multi-view stereo optical imagery for extraction of forest canopy height;