Advancing High-Resolution Surface Aboveground Biomass Modeling through Terrestrial Laser Scanning and Machine Learning in a Southeastern U.S. Pine Forest Ecosystem
- 1University of Florida, 1Forest Biometrics, Remote Sensing and Artificial Intelligence Laboratory (Silva Lab), School of Forest, Fisheries, and Geomatics Sciences, , (carine.klaubergs@ufl.edu)
- 2Instituto Nacional de Pesquisas Espaciais, Av. dos astronautas, 1758, São José dos Campos, 12227-010, Brazil;
- 3USDA Forest Service, Southern Research Station, Athens GA, USA;
- 4New Mexico Consortium, 4200 W. Jemez Rd., Los Alamos, NM 87544, USA;
- 5Rocky Mountain Research Station, USDA Forest Service, Forestry Sciences Laboratory, 1221 South Main Street, Moscow, ID 83843, USA
The integration of lidar (light detection and ranging) with machine learning offers a promising method for accurately estimating and mapping various vegetation attributes. This study demonstrated the effective use of terrestrial laser scanning (TLS) and the random forest (RF) machine learning approach to achieve precise total surface aboveground biomass (TSAGB) estimates at high resolution in regularly burned forest ecosystems in the southeastern United States. Our study site is located in the Osceola National Forest (ONF), which is part of the USFS Southern Research Station within the Olustee Experimental Forest, situated a short distance from Olustee, FL, and approximately 15 miles west of Lake City, FL. The site, spanning around 50 acres, was designated by the USFS Southeastern Forest & Range Experiment Station Southern Forest Fire Laboratory in 1957. Its primary purpose is to assess various fire return intervals and their impact on the accumulation of hazardous fuels. A total of 35 pre- burn clip plot (0.5m by 0.5m) samples in 2020 and 35 post-burn clip plot (0.5m by 0.5m) samples in 2022 were established for destructive TSAGB sampling. High-resolution 3D point cloud data were collected using the Riegl VZ 400i terrestrial laser scanner within each 5x5 meter collocated plot. A suite of cloud metrics was computed, and an RF model was fine-tuned for TSAGB, with a bootstrapping approach applied for model validation. The validation results indicated that a model utilizing only six metrics successfully predicted total TSAGB with relative and absolute Root Mean Square Error (RMSE) and bias of 57.59 g/m2 (32.78%) and -3.09 g/m2 (-1.76%), respectively. Our methodology, leveraging TLS lidar and widely used machine learning models, offers efficient solutions for enhancing the accuracy of surface biomass estimates in pine forests subject to frequent burns in the Southeastern U.S.
How to cite: Klauberg, C., da Silva, G. M., Loudermilk, E. L., Hawley, C. S., Pokswinski, S., Shimabukuro, Y. E., Sánchez-López, N., Hudak, A., and Silva, C. A.: Advancing High-Resolution Surface Aboveground Biomass Modeling through Terrestrial Laser Scanning and Machine Learning in a Southeastern U.S. Pine Forest Ecosystem, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13503, https://doi.org/10.5194/egusphere-egu24-13503, 2024.
