Forest Height Extraction Using GF-7 Very High-Resolution Stereoscopic Imagery and Google Earth Multi-Temporal Historical Imagery

• Published in: Journal of remote sensing 2024-01, Vol.4
• Main Authors: Ni, Wenjian, Li, Zijia, Wang, Qiang, Zhang, Zhiyu, Liu, Qingwang, Pang, Yong, He, Yating, Li, Zengyuan, Sun, Guoqing
• Format: Article
• Language: English
• ISSN: 2694-1589; 2694-1589
• DOI: 10.34133/remotesensing.0158

With the advent of very high-resolution (VHR) imaging satellites, it is possible to measure the heights of forest stands or even individual trees more accurately. However, the accurate geometric processing of VHR images depends on ground control points (GCPs). Collecting GCPs through fieldwork is time-consuming and labor-intensive, which presents great challenges for regional applications in remote or mountainous regions, particularly for international applications. This study proposes a promising approach that leverages GF-7 VHR stereoscopic images and Google Earth’s multi-temporal historical imagery to accurately extract forest heights without the need for fieldworks. Firstly, an algorithm is proposed to collect GCPs using Multi-temporal Averaging of historical imagery provided by Google Earth (GE), known as MAGE. Digital surface model (DSM) is then derived using GF-7 stereoscopic imagery and MAGE GCPs in Switzerland. Forest heights are finally extracted by subtracting ground surface elevations from GF-7 DSM. Results show that absolute coordinate errors of MAGE GCPs are less than 2.0 m. The root mean square error (RMSE) of forest heights extracted from GF-7 DSM, derived using the original geolocation model, is 12.3 m, and the determination coefficient ( R 2 ) of linear estimation model is 0.72. When the geolocation model is optimized using MAGE GCPs, the RMSE is reduced to 1.5 m and the R 2 increases to 0.95. These results not only demonstrate the effectiveness of MAGE GCPs but, more importantly, also reveal the significance of precise geometric processing of VHR stereoscopic imagery in forest height estimations.