SYSSIFOSS - Synthetic structural remote sensing data for improved forest inventory models

• Applicants: Professor Dr. Fabian Fassnacht; Professor Dr. Bernhard Höfle
• Subject Area: Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography ; Forestry
• Term: from 2019 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 411263134

• Final Report Year: 2023

Final Report Abstract

In the SYSSIFOSS project, an approach was developed to generate realistic synthetic laser scanning data of forests by combining forest growth simulators, a database of real tree point clouds and the laser scanning simulator HELIOS++. The suitability of the generated datasets to support laser scanning-based methods in the context of forest inventories was investigated. For the generation of the synthetic input data, airborne, UAV-borne, and terrestrial laser scanning data were collected in two forest areas in southwestern Germany, and single tree point clouds of 1491 trees were extracted from the data. The tree point clouds were published in a database along with metadata and metrics measured in the field and derived from the point clouds. The open-source Python package pytreedb was developed to enable easy access to the database in machine learning workflows and automatic data retrieval. The project investigated the required level of detail of the extracted tree point clouds to adequately represent the structure of individual trees in simulated data, and how laser scanning simulations must be configured to achieve sufficient realism. For this purpose, simulated and real laser scanning point clouds of single trees as well as of six 1-ha forest plots were compared using different point cloud metrics. The investigations at the individual tree level showed that a voxel model with very small voxels (2 cm side length) or with voxels scaled according to the local leaf area density is best suited to convert the tree point clouds into scannable object geometries for TLS simulations. The area-based comparison showed that, depending on the application, simplified tree models are also suitable for generating synthetic laser scanning data. Another aim of the project was to test the suitability of the synthetic data for improving forest inventory workflows based on laser scanning data. A sensitivity analysis on the simulated data was conducted to investigate how the pulse repetition frequency, the inventory method for collecting reference data, and a co-registration error between laser scanning data and reference data affect regression models for estimating forest aboveground biomass. The ANOVA revealed that the model quality particularly depends on the field inventory method, while the pulse repetition frequency and the co-registration error have less influence on the model quality. Furthermore, it was examined to what extent it is possible to replace field measurements for the calibration of laser scanning-based forest inventory models with synthetic data. It was found that biomass models calibrated exclusively with synthetic data have a significantly lower accuracy compared to models that use real field data for calibration (increase of the RMSE by 26-105%, depending on the study site). However, the results showed that synthetic data are well suited as supplementary calibration data when only few field measurements are available. This indicates that synthetic data can be used to reduce the number of field plots while maintaining the same model accuracy. The 3D tree point clouds and accompanying field- and meta-data collected in this project are accessible as open data and can be used for a variety of studies, also beyond this project, as first publications by other research groups show. During the project, the open-source HELIOS++ laser scanning simulation software was extended in both functionality and performance and has developed into an often employed and cited research software.

Publications

• Opaque voxel-based tree models for virtual laser scanning in forestry applications. Remote Sensing of Environment, 265 (2021, 11), 112641.
  Weiser, Hannah; Winiwarter, Lukas; Anders, Katharina; Fassnacht, Fabian Ewald & Höfle, Bernhard
  
• Individual tree point clouds and tree measurements from multi-platform laser scanning in German forests. Earth System Science Data, 14(7), 2989-3012.
  Weiser, Hannah; Schäfer, Jannika; Winiwarter, Lukas; Krašovec, Nina; Fassnacht, Fabian E. & Höfle, Bernhard
  
• Virtual laser scanning with HELIOS++: A novel take on ray tracing-based simulation of topographic full-waveform 3D laser scanning. Remote Sensing of Environment, 269, 112772.
  Winiwarter, Lukas; Esmorís, Pena Alberto Manuel; Weiser, Hannah; Anders, Katharina; Martínez, Sánchez Jorge; Searle, Mark & Höfle, Bernhard
  
• Virtual LiDAR Simulation as a High Performance Computing Challenge: Toward HPC HELIOS++. IEEE Access, 10(2022), 105052-105073.
  Esmoris, Alberto M.; Yermo, Miguel; Weiser, Hannah; Winiwarter, Lukas; Hofle, Bernhard & Rivera, Francisco F.
  
• Generating synthetic laser scanning data of forests by combining forest inventory information, a tree point cloud database and an open-source laser scanning simulator. Forestry: An International Journal of Forest Research, 96(5), 653-671.
  Schäfer, Jannika; Weiser, Hannah; Winiwarter, Lukas; Höfle, Bernhard; Schmidtlein, Sebastian & Fassnacht, Fabian Ewald