Estimating Aboveground Biomass of North American Forests Using a Three-Phased Sampling Approach Based on Ground Plots, Airborne Lidar, and the GLAS Satellite Sensor

Hank A. Margolis, Laval University, hank.a.margolis@nasa.gov (Presenter)
Ross Nelson, NASA GSFC, ross.f.nelson@nasa.gov
Paul Montesano, NASA/SSAI, paul.m.montesano@nasa.gov
Guoqing Sun, NASA GSFC/UMD, guoqing.sun@nasa.gov
André Beaudoin, Canadian Forest Service, abeaudoin@exchange.cfl.forestry.ca
Hans Erik Andersen, USDA Forest Service, handersen@fs.fed.us
Michael Wulder, Canadian Forest Service, mike.wulder@nrcan-rncan.gc.ca
Bruce Cook, NASA GSFC, bruce.cook@nasa.gov
Lawrence A Corp, SSAI, lawrence.a.corp@nasa.gov
Bernardus de Jong, El Colegio de la Frontera Sur (ECOSUR), bjong@ecosur.mx
Fernando Paz, Colegio De Postgraduados, pellat@colpos.mx

We implemented a three-phase sampling design to estimate aboveground forest biomass for North American forests that could help serve as a proof of concept for a Decadal Survey satellite mission to use lidar sampling to calculate regional biomass. We sampled ground plots with an airborne lidar system, and then used the same airborne lidar to fly over the ground tracks of lidar pulses from the Geosciences Lidar Altimeter System (GLAS) on board the ICESat satellite. Airborne campaigns were conducted in Quebec in 2005, Alaska in 2008, rest of Canada in 2009, eastern US in 2011, western US in 2012, and Mexico in 2013. Biomass estimates were stratified by ecozone and forest cover type. Regressions between ground plot biomass and airborne lidar metrics were strong in all regions, e.g., R2 from 0.51 to 0.86 (RMSE = 12 to 56 Mg/ha) in Mexican forests and from 0.50 to 0.84 (RMSE = 15.1 to 33.2 Mg/ha) in boreal forests. The weakest regressions were for non-conifer cover types in Central Mexican Dry Forests, likely due to variations in canopy phenology at the time of the sampling. In tropical regions, Yucatan moist hardwoods had an R2 of 0.57 (RMSE = 56 Mg/ha) while the Yucatan dry hardwoods had R2 = 0.86 (RMSE = 24 Mg/ha). Subsequent regressions between airborne-derived biomass and GLAS metrics were sufficiently strong (e.g., R2 = 0.52 to 0.79, RMSE = 9.9 to 29.6 Mg/ha for boreal forests) to allow us to use GLAS as a regional sampling tool. Uncertainty estimates were divided into sampling error and airborne-GLAS model error. The percent model error across ecozones was related to the number of GLAS orbits crossing a given ecozone as well as with the number of GLAS pulses. Small ecozones oriented primarily in a north-south direction had the greatest percent sampling error. We estimated 21.8 ±4.1 Pg of aboveground biomass for the boreal forest of North America with 9.7% in Alaska, 46.6% in western Canada, and 43.7% in eastern Canada. Overall, 51.3% of the boreal biomass was in conifers, 22.0% in mixedwoods, 14.3% in hardwoods, 11.4% in forested wetlands, and 1.1% in recent burns. Our GLAS-derived boreal forest biomass estimates were consistent with those derived from Canada’s National Forest Inventory system and with those derived from a kNN modeling approach based on MODIS data.

Presentation Type:  Poster

Session:  Theme 3: Future research direction and priorities: perspectives relevant to the next decadal survey   (Mon 4:30 PM)

Associated Project(s): 

• Nelson, Ross: A Lidar-Radar-Optical Data Fusion Approach for Estimating the Aboveground Carbon Stocks of North American Forests: Means and Uncertainties at Regional to Continental Scales ...details
• Nelson, Ross: Using the ICESAT-GLAS LiDAR to Estimate the Amount, Spatial Distribution, and Statistical Uncertainty of Aboveground Carbon Stocks of the North American Boreal Forest ...details

Poster Location ID: 206