The Chinese Academy of Astronautics pioneered a new ultra-high resolution optical forest 3D remote sensing method

Recently, the forest remote sensing team led by Ni Wenjian, a researcher at the State Key Laboratory of Remote Sensing Science of the Academy of Aerospace Information Innovation of the Chinese Academy of Sciences, has made important progress in the study of extracting the three-dimensional structure of forests using ultra-high-resolution optical remote sensing stereoscopic observation data. According to the existing research, the optical multi-angle stereoscopic observation data does not have the penetrating ability in the forest area, so the direct measurement of forest vertical structure parameters cannot be carried out independently in the absence of understory topographic data, especially in the dense mountain forest area. It is found that the optical stereoscopic observation data with a resolution of better than 0.2 m can accurately characterize the crown structure of individual trees. Inspired by the allometric growth equation of trees, we created the Growth Relation-Constrained Understory Terrain Approximation Algorithm (AGAR), which breaks the traditional cognitive limitation and realizes the direct measurement of forest vertical structure using only optical stereoscopic observation data. The research is in the journal Remote Sensing of Environment.

As one of the important carbon pools in terrestrial ecosystem, the accurate estimation of forest carbon storage is the main direction of remote sensing research, which can serve China's "double carbon" strategy and the study of Earth system carbon cycle process. In the past, the theory and method of forest carbon storage estimation based on "two-dimensional" characteristics such as remote sensing image spectrum or microwave scattering intensity have been studied at home and abroad, but "terrain influence" and "remote sensing signal saturation" are still two major scientific problems that are difficult to overcome. Therefore, the international community has gradually turned to "three-dimensional" remote sensing based on satellite ranging technology, including LiDAR remote sensing based on laser ranging, synthetic aperture radar interference based on microwave ranging, and optical multi-angle stereoscopic observation based on visual ranging. Scientists in the United States are working to develop space-borne lidar with canopy penetration, These include the early laser altimeters SLA01 and SLA02 that flew on the Space shuttle, the ICESat/GLAS satellites that operated from 2003 to 2009, the ICESAT-2 satellite that launched in 2018, and the GEDI that will be placed on the International Space Station in 2019. European researchers are actively developing L-band TANTANDEM -L and P-band BIOMASS synthetic aperture radar interference satellites with strong penetration capacity, and plan to launch in 2024. Compared with laser radar and synthetic aperture radar interferometry, optical multi-angle stereo remote sensing has a significant advantage of visual image, but is limited by the penetration ability. Currently, it is mainly used for the measurement of surface elevation, and it needs to rely on the understory topography provided by other data sources to measure the vertical structure of forests, and its application value and scenarios are limited.

In recent years, China has made rapid progress in optical multi-angle stereo remote sensing, and has launched Yuan-3, Gaofen-7, Tianhui series and other commercial remote sensing satellites, while the spatial resolution of images has gradually improved. Whether the continuous improvement of spatial resolution can be used to break through the limitation of its weak penetration ability, and then maximize the application value of ultra-high resolution optical multi-angle stereo remote sensing data is not only an international frontier scientific problem, but also an urgent question for Chinese remote sensing researchers.

The forest remote sensing team realized the unique value of ultra-high resolution optical multi-angle stereoscopic observation data. Since 2014, it has carried out continuous research on the application of UAV stereoscopic observation data in the measurement of forest structure parameters. In 2018, it carried out a large-scale UAV sampling observation experiment in the Greater Hinggan Mountains forest area, revealing the coupling law of observation Angle and image resolution. The complementary effect of forest height information on the estimation of leaf area index was confirmed, and a collaborative image solution for forest height extraction with and without leaves was developed for deciduous forest area, which broke through the identification technology of scattered dead trees, single tree identification and segmentation technology, and high-precision forest cover extraction technology based on background recognition. On the basis of the above data and technology accumulation, the team created the "Growth-relationship-constrained Understory Terrain Approximation Algorithm" (AGAR) to achieve direct extraction of forest height under complex terrain conditions. This result confirms that the AGAR algorithm can extract forest height using only ultra-high resolution optical multi-angle stereoscopic observation data without the support of additional understory topographic data.

Although the AGAR algorithm uses stereoscopic images acquired by drones for research, and the specific technical details of the algorithm need to be further tested and refined, with the advent of 0.1 m satellite optical remote sensing data, the method will open a new era of ultra-high resolution optical stereo remote sensing forest 3D remote sensing.

Figure 1. The core idea of growth relation constrained understory terrain approximation algorithm (AGAR)

Figure 2. Effect of forest height extraction under typical terrain conditions. (a) - (c) Digital surface model (DSM) for optical multi-angle stereoscopic data; (d) - (f) refers to the forest height extracted from optical multi-angle stereoscopic observation data through forest window interpolation. Due to the small number of forest Windows in dense forest areas, tree height is seriously underestimated or topographic features are not completely removed; (g) - (i) refers to the height of the forest extracted using AGAR. (a) The area covers the ridge and (b) the area covers the valley; (c) The area covers the slope from the base of the mountain to the summit.

Source: Institute of Aerospace Information Innovation, Chinese Academy of Sciences