Accuracy evaluation and effect factor analysis of GEDI aboveground biomass product for temperate forests in the conterminous United States

Figures & data

Figure 1. Distribution of the 19 NEON sites and their corresponding forest types across the CONUS. Italic numbers represent the number of eco-climate domains which correspond to the IDs in table S1 in supplementary.

Table 1. Geographic location and environmental characteristics of the NEON sites.

Site	Eco-climatic domain	Altitude (m)	Area (km^2)	Forest type
ABBY	16	363	74.48	Broadleaf forest
BART	1	232	118.35	Old-growth deciduous forest
BLAN	2	183	224.44	Fragmented secondary forests and successional shrubs
CLBJ	11	272	158.77	Forest and grassland
DELA	8	25	122.15	Forest and agriculture
GRSM	7	579	166.12	Deciduous old growth forest
HARV	1	348	345.20	Deciduous forest
LENO	8	20	114.84	Mixed forest, grassland and wetlands
MLBS	7	1126	165.57	Deciduous forest
NIWO	13	3513	152.16	Mixed forest
RMNP	10	2743	189.67	Evergreen coniferous forest
SERC	2	15	245.35	Mixed forest
SOAP	17	1210	172.01	Evergreen Forest, Shrub/Scrub
STEI	5	481	296.58	Deciduous Forest, Mixed Forest, Woody Wetlands
TALL	8	166	136.16	Deciduous Forest, Evergreen Forest, Mixed Forest
TEAK	17	2147	188.17	Evergreen Forest, Shrub/Scrub
UKFS	6	322	191.78	Evergreen coniferous forest
WREF	16	351	395.91	Evergreen coniferous forest
YELL	12	2116	245.24	Evergreen Forest, Grassland/Herbaceous, Shrub/Scrub

Note: Area refers to the area covered by the airborne Lidar in each NEON site which is described in Section 2.2.2.

Table 2. Summary of screening strategies for GEDI L2A.

Screening strategy	Justification
quality_flag = 1	Selection of high-quality footprints
degrade_flag = 0
landsat_water_persistence > 10%	Avoidance of permanent water
urban proportion greater than 0	Avoidance of urban footprints
leaf_off_flag = 0	Selection of footprints in growing season
elevation difference between the GEDI and the TanDEM-X DEM < 50 m	Avoidance of elevation error
landsat_treecover > 0	Selection of forested footprints
modis_treecover > 0
vegetation height > 5 m

Table 3. Five raster AGBD products.

Maps	Resolution	Input	Method	Reference year
Yang et al. (2020)	1 km	LiDAR-derived biomass datasets and field measurements, high-level products from remote sensing data (LAI, GPP, VCF, and land cover type), canopy height, climate variables, and topographic data	Gradient Boosting Decision Tree	2005
Santoro et al. (2021)	100 m	ALOS PALSAR, Envisat ASAR, Landsat, GLAS	BIOMASAR, allometric equation	2010
Hu et al. (2016)	1 km	Forest ground inventory, GLAS, Normalized Difference MODIS vegetation index, topographic, climate, land cover, slope, GLAS waveform	Random Forest	1990 ~ 2013
Wilson et al. (2013)	250 m	FIA field data, MODIS vegetation indices time series, monthly mean climatic characteristics, topographic metrics, ecological zones	Nearest-neighbor imputation approach	2000–2009
Thurner et al. (2014)	0.01°	growing stock volume by BIOMASAR from SAR data, Global Wood Density Database, JRC GHG-AFOLU Biomass Compartment Database	BIOMASAR Allometric equation	2010

Table 4. Categories and data sources of the selected nine factors.

Number	Category	Effect factors	Data source
1	SWSD	SWSD	GEDI L2A and NEON Lidar
2	Canopy characteristics	Tree height	NEON individual tree products
		Crown size	NEON individual tree products
		Canopy cover	GEDI L2A
3	Canopy heterogeneity	Tree density	NEON individual tree products
		Crown size SD	NEON individual tree products
		Tree height SD	NEON individual tree products
4	Other factors	Forest type	Forest type inventory
		Topographic slope	NEON DTMs

Figure 2. Flowchart of methodology of GEDI L4A accuracy evaluation and uncertainty estimation.

Table 5. Accuracy of NEON lidar AGBD for the different NEON sites.

Site	Plot number	RMSE (Mg/ha)	MAE (Mg/ha)	%RMSE	ubRMSE
ABBY	22	75.05	93.14	47.89%	60.74
BART	22	42.67	37.75	17.29%	35.27
BLAN	16	42.27	36.27	26.11%	29.83
CLBJ	27	33.17	27.44	31.65%	29.61
DELA	29	47.23	38.06	24.05%	42.43
GRSM	15	44.00	43.55	18.05%	42.63
HARV	29	77.49	67.16	48.88%	76.14
LENO	24	111.10	85.13	29.67%	87.03
MLBS	29	51.80	36.69	24.47%	35.79
NIWO	20	34.59	27.75	23.89%	23.23
RMNP	22	30.96	24.06	26.20%	26.23
SERC	30	80.69	57.33	27.74%	62.92
SOAP	28	54.77	46.51	39.78%	50.91
STEI	15	20.44	18.04	42.40%	12.62
TALL	25	50.50	42.88	36.55%	43.95
TEAK	20	67.24	58.17	24.94%	54.05
UKFS	27	56.34	46.14	16.44%	49.52
WREF	24	98.72	79.37	34.89%	76.60
YELL	19	27.46	21.61	26.50%	21.02

Figure 3. Comparison of GEDI L4A’s accuracy among NEON sites. Data points in scatter plots represents GEDI footprints.

Figure 4. Comparison of GEDI L4A’s accuracy among forest types. The number of each forest type corresponds to the one in Figure 1. Data point in the scatter plots represents GEDI footprints.

Figure 5. Spatial distribution of GEDI L4B’s uncertainty and relative uncertainty.

Table 6. Statistics of GEDI L4B’s uncertainty and relative uncertainty.

Eco-climatic domains	Corresponding NEON sites	Mean value of uncertainty (Mg/ha)	Mean value of relative uncertainty
Northeast	BART, HARV	41.14	0.30
Mid Atlantic	BLAN, SERC	45.29	0.34
Southeast	-	35.22	0.37
Atlantic Neotropical	-	18.34	0.55
Great Lakes	STEI	34.58	0.36
Prairie Peninsula	UKFS	44.75	0.30
Appalachians/Cumberland Plateau	MLBS, GRSM	54.73	0.30
Ozarks Complex	TALL, DELA, LENO	41.88	0.34
Northern Plains	-	23.06	0.33
Central Plains	-	21.79	0.46
Southern Plains	CLBJ	42.28	0.40
Northern Rockies	YELL	35	0.37
Southern Rockies/Colorado Plateau	NIWO, RMNP	24.52	0.33
Desert Southwest	-	40.46	0.37
Great Basin	-	31.09	0.36
Pacific Northwest	WREF, ABBY	77.78	0.38
Pacific Southwest	SOAP, TEAK	50.28	0.35

Figure 6. Relationship between different effect factors and the AD of GEDI L4A. Δ|Bias| represents the maximum difference of the mean value of each group.

Table 7. Relative importance of factors to the AD of GEDI L4A.

Category	Relative importance (%)	Effect factors	Relative importance (%)
SWSD	56.63	SWSD	56.63
Canopy characteristics	3.99	Tree height	3.27
		Crown size	0.55
		Canopy cover	0.17
Canopy heterogeneity	32.40	Tree density	0.86
		Crown size SD	8.71
		Tree height SD	22.83
Other factors	6.98	Forest type	6.27
		Topographic slope	0.71

Figure 7. Structural equation model between the AD of GEDI L4A and multiple factors with significant correlation.

Table 8. Relative importance of factors to SWSD.

Category	Relative importance (%)	Effect factors	Relative importance (%)
Canopy characteristics	33.78	Tree height	4.50
		Crown size	20.87
		Canopy cover	8.41
Canopy heterogeneity	54.52	Tree density	40.42
		Crown size SD	0.95
		Tree height SD	13.15
Other factors	11.70	Forest type	11.67
		Topographic slope	0.03

Figure 8. Comparison of the spatial distribution of GEDI L4B with 5 other AGBD products.

Figure 9. Comparison of a statistical histogram of GEDI L4B with other 5 AGBD products across the CONUS.

Yang, L., S. Liang, and Y. Zhang. 2020. “A New Method for Generating a Global Forest Aboveground Biomass Map from Multiple High-Level Satellite Products and Ancillary Information.” IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing 13:2587–2597. https://doi.org/10.1109/JSTARS.2020.2987951.

 Web of Science ®Google Scholar

Santoro, M., O. Cartus, N. Carvalhais, D. M. A. Rozendaal, V. Avitabile, A. Araza, S. de Bruin, et al. 2021. “The Global Forest Above-Ground Biomass Pool for 2010 Estimated from High-Resolution Satellite Observations.” Earth System Science Data 13 (8): 3927–3950. https://doi.org/10.5194/essd-13-3927-2021.

 Web of Science ®Google Scholar

Hu, T., Y. Su, B. L. Xue, J. Liu, X. Zhao, J. Fang, and Q. Guo. 2016. “Mapping Global Forest Aboveground Biomass with Spaceborne LiDar, Optical Imagery, and Forest Inventory Data.” Remote Sensing 8 (7): 565. https://doi.org/10.3390/rs8070565.

 Web of Science ®Google Scholar

Wilson, B. T., C. W. Woodall, and D. M. Griffith. 2013. “Imputing Forest Carbon Stock Estimates from Inventory Plots to a Nationally Continuous Coverage.” Carbon Balance and Management 8 (1): 1. https://doi.org/10.1186/1750-0680-8-1.

 PubMedGoogle Scholar

Thurner, M., C. Beer, M. Santoro, N. Carvalhais, T. Wutzler, D. Schepaschenko, A. Shvidenko, et al. 2014. “Carbon Stock and Density of Northern Boreal and Temperate Forests.” Global Ecology & Biogeography 23 (3): 297–310. https://doi.org/10.1111/geb.12125.

 Web of Science ®Google Scholar

Data availability statement

The raw discrete point cloud LiDAR and GEDI data were generated at the National Ecological Observation Network (NEON) and the Land Processes Distributed Active Archive Center (LP DAAC), respectively. Derived data supporting the findings of this study are available from the corresponding author on request.