Annual aboveground biomass mapping for North America’s Boreal Forest during 1984-2022 with ground plots, airborne lidar, and Landsat time series
Wanwan Liang, University of Utah, u6054045@umail.utah.edu (Presenter)
Kai-Ting Hu, Boston University, kthu@bu.edu
Piotr Tompalski, Canadian Forest Service, piotr.tompalski@nrcan-rncan.gc.ca
Mark A. Friedl, Boston University, friedl@bu.edu
James T. Randerson, University Of California, Irvine, jranders@uci.edu
Olivier R. van Lier, Canadian Forest Service, olivier.vanlier@nrcan-rncan.gc.ca
Douglas C. Morton, NASA GSFC, douglas.morton@nasa.gov
Hans Erik Andersen, U.S. Forest Service Pacific Northwest Research Station, handersen@fs.fed.us
Jonathan A. Wang, University of Utah, jon.wang@utah.edu
Background: Accurate time series maps of aboveground biomass (AGB) are crucial for characterizing how the carbon cycle is responding to rapid climate change and increasing disturbance like wildfire. Regionally consistent field-calibrated maps of annual AGB with high spatial resolution are currently unavailable for the entire North American boreal zone, but are necessary to characterize carbon dynamics and compare with estimates from models and atmospheric inversions. Therefore, our primary goal is to generate spatially exhaustive annual AGB maps for the North American Arctic and Boreal regions from 1984 to 2022.
Methods: To calibrate and validate the annual AGB maps, we have compiled and analyzed 2,171 ground plots from various forest inventory programs and more than 116,000 km² of airborne laser scanning (ALS) data from multiple programs in Canada and Goddard’s Lidar, Hyperspectral and Thermal Imager (G-LiHT) program in Alaska. We used a two-step modeling approach. Using the area-based approach we first derived localized maps of AGB from the ALS data calibrated with spatially corresponding ground plots. To scale these observations across space and time, we analyzed over 100TB of Landsat Collection 2 Surface Reflectance using the Continuous Change Detection and Classification (CCDC) algorithm to create seasonally corrected synthetic reflectance values and develop spectral-temporal input features. These features and ancillary data (i.e., land cover, topographical variables, and long-term mean climate) were then used in the second modeling step as predictors in a machine learning model calibrated with the ALS-derived AGB predictions to generate preliminary wall-to-wall 30 m resolution maps of annual AGB. We then used ground-based AGB from repeated field surveys and canopy height from repeated ALS acquisitions to analyze whether Landsat-derived AGB can capture the temporal changes in forest attributes.
Preliminary results: We observed high accuracy for both the model that used the ground plots to calibrate AGB model with ALS (R-square = 0.82, MAE = 30 Mg/ha), and the model that used the ALS-derived AGB to calibrate AGB model with Landsat data (R-square = 0.90, MAE = 19 Mg/ha). Validation of the preliminary model using independent ground plots (R-square > 0.40, MAE < 50 Mg/ha) showed the best accuracy based on R-square, and the second-best accuracy based on MAE compared to three existing AGB maps partially covering the study domain (R-square: 0.29-0.40, MAE: 47-54 Mg/ha). In all maps analyzed, there was a consistent underestimation of AGB at high AGB levels, with some maps also showing underestimation at relatively low AGB levels. AGB change analysis based on our preliminary map indicated that Landsat-derived AGB can capture the sign of temporal AGB and canopy height change in more than 65% and 80% of cases, respectively.
Next steps: To improve AGB maps, we will integrate newly acquired ground plots (>25,000) for model calibration and synthesizing additional airborne data from LVIS and spaceborne LiDAR from ICESat-2. This comprehensive analysis is expected to significantly contribute to the advancement of AGB mapping techniques and enable robust quantification of the impacts of climate change and disturbance on the Arctic-boreal carbon cycle.
Poster: Poster_Liang_13_51_41.pdf
Associated Project(s):
Poster Location ID: 13
Presentation Type: Poster
Session: Vegetation Structure and Function
Session Date: Tuesday (5/21) 5:00 PM