GEDI L4A Footprint Level Aboveground Biomass Density, Version 2

This dataset contains Global Ecosystem Dynamics Investigation (GEDI) Level 4A (L4A) Version 2 predictions of the aboveground biomass density (AGBD; in Mg/ha) and estimates of the prediction standard error within each sampled geolocated laser footprint. In this version, the granules are in sub-orbits. The algorithm setting group selection used for GEDI02_A Version 2 has been modified for Evergreen Broadleaf Trees in South America to reduce false positive errors resulting from the selection of waveform modes above ground elevation as the lowest mode. The footprints are located within the global latitude band observed by the International Space Station (ISS), nominally 51.6 degrees N and S and reported for the period 2019-04-18 to 2021-08-05. The GEDI instrument consists of three lasers producing a total of eight beam ground transects, which instantaneously sample eight ~25 m footprints spaced approximately every 60 m along-track. The GEDI beam transects are spaced approximately 600 m apart on the Earth's surface in the cross-track direction, for an across-track width of ~4.2 km. Footprint AGBD was derived from parametric models that relate simulated GEDI Level 2A (L2A) waveform relative height (RH) metrics to field plot estimates of AGBD. Height metrics from simulated waveforms associated with field estimates of AGBD from multiple regions and plant functional types (PFTs) were compiled to generate a calibration dataset for models representing the combinations of world regions and PFTs (i.e., deciduous broadleaf trees, evergreen broadleaf trees, evergreen needleleaf trees, deciduous needleleaf trees, and the combination of grasslands, shrubs, and woodlands). For each of the eight beams, additional data are reported with the AGBD estimates, including the associated uncertainty metrics, quality flags, model inputs, and other information about the GEDI L2A waveform for this selected algorithm setting group. Model inputs include the scaled and transformed GEDI L2A RH metrics, footprint geolocation variables and land cover input data including PFTs and the world region identifiers. Additional model outputs include the AGBD predictions for each of the six GEDI L2A algorithm setting groups with AGBD in natural and transformed units and associated prediction uncertainty for each GEDI L2A algorithm setting group. Providing these ancillary data products will allow users to evaluate and select alternative algorithm setting groups. Also provided are outputs of parameters and variables from the L4A models used to generate AGBD predictions that are required as input to the GEDI04_B algorithm to generate 1-km gridded products. Note that there are 351 granules in this release affected by duplicate GEDI shots for selected days (2020-297 to 2020-300, 2020-365, and 2021-106).

There are 44,095 data files in HDF5 (*.h5) format included in this dataset and four companion files that provide additional details regarding the product model development and variable descriptions. Companion files must be downloaded separately from the dataset.

This dataset contains Global Ecosystem Dynamics Investigation (GEDI) Level 4A (L4A) Version 2 predictions of the aboveground biomass density (AGBD; in Mg/ha) and estimates of the prediction standard error within each sampled geolocated laser footprint. The granules in this version 2 dataset are in sub-orbits. The algorithm setting group selection used for GEDI02_A Version 2 has been modified for Evergreen Broadleaf Trees in South America to reduce false-positive errors resulting from the selection of waveform modes above ground elevation as the lowest mode. The footprints are located within the global latitude band observed by the International Space Station (ISS), nominally 51.6 degrees N and S, and reported for the period 2019-04-17 to 2021-08-05. The GEDI instrument consists of three lasers producing a total of eight beam ground transects, which instantaneously sample eight ~25 m footprints spaced approximately every 60 m along-track. The GEDI beam transects are spaced approximately 600 m apart on the Earth's surface in the cross-track direction, for an across-track width of ~4.2 km. Footprint AGBD was derived from parametric models that relate simulated GEDI Level 2A (L2A) waveform relative height (RH) metrics to field plot estimates of AGBD. Height metrics from simulated waveforms associated with field estimates of AGBD from multiple regions and plant functional types (PFT) were compiled to generate a calibration dataset for models representing the combinations of world regions and PFTs (i.e., deciduous broadleaf trees, evergreen broadleaf trees, evergreen needleleaf trees, deciduous needleleaf trees, and the combination of grasslands, shrubs, and woodlands).

Uncertainty metrics, quality flags, and model inputs are reported with the AGBD estimates for each of the eight beams. Model inputs include the scaled and transformed GEDI L2A RH metrics and other information about the GEDI L2A waveform for this selected algorithm setting group. Also provided are model inputs for each of the eight beams including footprint geolocation variables, land cover input data including PFTs, and the world region identifiers. Additional model outputs include the AGBD predictions for each of the six GEDI L2A algorithm setting groups with AGBD in natural and transformed units and associated prediction uncertainty. These ancillary data products allow users to evaluate and select alternative algorithm setting groups. The outputs of parameters and variables from the L4A models used to generate AGBD predictions are also provided; these outputs serve as input to the GEDI04_B algorithm to generate 1-km gridded products.

Project: Global Ecosystem Dynamics Investigation

The Global Ecosystem Dynamics Investigation (GEDI) produces high resolution laser ranging observations of the 3D structure of the Earth. GEDI’s precise measurements of forest canopy height, canopy vertical structure, and surface elevation greatly advance our ability to characterize important carbon and water cycling processes, biodiversity, and habitat. GEDI was funded as a NASA Earth Ventures Instrument (EVI) mission. It was launched to the International Space Station in December 2018 and completed initial orbit checkout in April 2019.

Related Publication

Dubayah, R., J.B. Blair, S. Goetz, L. Fatoyinbo, M. Hansen, S. Healey, M. Hofton, G. Hurtt, J. Kellner, S. Luthcke, J. Armston, H. Tang, L. Duncanson, S. Hancock, P. Jantz, S. Marselis, P.L. Patterson, W. Qi, and C. Silva. 2020. The Global Ecosystem Dynamics Investigation: High-resolution laser ranging of the Earth’s forests and topography. Science of Remote Sensing 1:100002. https://doi.org/10.1016/j.srs.2020.100002

Duncanson, L., Kellner, J.R., Armston, J., Dubayah, R., Minor, D.M., Hancock, S., Healey, S.P., Patterson, P.L., Saarela, S., Marselis, S. and Silva, C.E., 2022. Aboveground biomass density models for NASA’s Global Ecosystem Dynamics Investigation (GEDI) lidar mission. Remote Sensing of Environment, 270, p.112845.  https://doi.org/10.1016/j.rse.2021.112845

Related Datasets

Dubayah, R.O., J. Armston, J.R. Kellner, L. Duncanson, S.P. Healey, P.L. Patterson, S. Hancock, H. Tang, M.A. Hofton, J.B. Blair, and S.B. Luthcke. 2022. GEDI L4A Footprint Level Aboveground Biomass Density, Golden Weeks, Version 1. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/2028

Dubayah, R.O., J. Armston, J.R. Kellner, L. Duncanson, S.P. Healey, P.L. Patterson, S. Hancock, H. Tang, M.A. Hofton, J.B. Blair, and S.B. Luthcke. 2021. GEDI L4A Footprint Level Aboveground Biomass Density, Version 1. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1907

Dubayah, R.O., S.B. Luthcke, T.J. Sabaka, J.B. Nicholas, S. Preaux, and M.A. Hofton. 2021. GEDI L3 Gridded Land Surface Metrics, Version 2. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1952

Dubayah, R.O., S.B. Luthcke, T.J. Sabaka, J.B. Nicholas, S. Preaux, and M.A. Hofton. 2021. GEDI L3 Gridded Land Surface Metrics, Version 1. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1865

Level 1B, Level 2A, and Level 2B data from GEDI are available from the Land Processes Distributed Active Archive Center at https://lpdaac.usgs.gov/

Acknowledgments

This work was funded by NASA contract #NNL 15AA03C to the University of Maryland for the development and execution of the GEDI mission (Dubayah, Principal Investigator). We thank the NASA Terrestrial Ecology Program and Hank Margolis for supporting the GEDI mission, and the University of Maryland for providing independent financial support. We thank Jamis Bruening, Suzanne Marselis, David Minor, and Carlos E. Silva for contributing to the development and management of the GEDI Forest Structure and Biomass Database. We gratefully acknowledge the GEDI Science Team and numerous collaborators who generously contributed field estimates of AGBD, stem maps, and airborne lidar data. These people include Katharine Abernethy, Hans-Erik Andersen, Paul Aplin, Timothy R. Baker, Nicolas Barbier, Jean Francois Bastin, Pascal Boeckx, Jan Bogaert, Luigi Boschetti, Peter Brehm Boucher, Doreen S. Boyd, Patrick Burns, David F.R.P. Burslem, Sofia Calvo-Rodriguez, Jérôme Chave, Robin L. Chazdon, David B. Clark, Deborah A. Clark, Warren B. Cohen, David A. Coomes, Piermaria Corona, K.C. Cushman, Mark E. J. Cutler, James William Dalling, Michele Dalponte, Sergio de-Miguel, Songqiu Deng, Peter Woods Ellis, Barend Erasmus, Michael Falkowski, Temilola Fatoyinbo, Patrick A. Fekety, Alfredo Fernández-Landa, Antonio Ferraz, Rico Fischer, Adrian G. Fisher, Antonio García-Abril, Terje Gobakken, Scott J. Goetz, Jonathan A. Greenberg, Jorg M. Hacker, Matt Hansen, Marco Heurich, Ross A. Hill, Sören Holm, Chris Hopkinson, Chengquan Huang, Huabing Huang, Stephen P. Hubbell, Andrew T. Hudak, George Hurtt, Andreas Huth, Benedikt Imbach, Patrick Jantz, Kathryn Jeffery, Masato Katoh, Elizabeth Kearsley, Natascha Kljun, Nikolai Knapp, Kamil Král, Martin Krucek, Nicolas Labrière, Seung-kuk Lee, Simon L. Lewis, Marcos Longo, Richard M. Lucas, Scott Luthcke, Russell Main, Jose A. Manzanera, Suzanne Marselis, Rodolfo Vásquez Martínez, Renaud Mathieu, Victoria Meyer, Paul Montesano, Felix Morsdorf, Erik Næsset, Laven Naidoo, Reuben Nilus, Michael J. O'Brien, David A. Orwig, Geoffrey Parker, Paul Patterson, Christopher Philipson, Oliver L. Phillips, Jan Pisek, Jim Pontius, John R. Poulsen, Wenlu Qi, Christoph Rüdiger, Svetlana Saarela, Sassan Saatchi, Arturo Sanchez-Azofeifa, Nuria Sanchez-Lopez, Crystal B. Schaff, Marc Simard, Andrew Kerr Skidmore, Göran Ståhl, Krzysztof Sterenczak, Chiara Torresan, Rubén Valbuena, Hans Verbeeck, Tomas Vrska, Konrad Wessels, Joanne C. White, and Carlo Zgraggen.

Spatial Resolution: Footprints ~25 m in diameter

Temporal Coverage: 2019-04-17 to 2021-08-05

Temporal Resolution: One-time estimate

Study Area: Latitude and longitude are given in decimal degrees.

Site	Westernmost Longitude	Easternmost Longitude	Northernmost Latitude	Southernmost Latitude
Global	-180	180	53.99333	-52.20456

Data File Information

There are 44,095 data files in HDF5 (*.h5) format included in this dataset. Each file provides multiple datasets/groups for each of the eight beams with valid data (­­i.e., 0000, 0001, 0010, 0011, 0101, 0110, 1000, 1011). There are also four companion files that provide additional details regarding the product model development and variable descriptions. Companion files must be downloaded separately from the dataset.

The files are named GEDI04_A_YYYYDDDHHMMSS_O[orbit_number]_[granule_number]_T[track_number]_[PPDS_type]_ [release_number]_[production_version]_V[version_number].h5 (e.g., GEDI04_A_2021188232338_O14550_04_T08520_02_002_01_V002.h5), where:

GEDI04_A = product short name representing GEDI Level 4A data,
YYYYDDDHHMMSS = date and time of acquisition in Julian day of year, hours, minutes, and seconds format,
[orbit_number] = orbit number,
[granule_number] = sub-orbit granule (or file) number,
[track_number] = track number,
[PPDS_type] = positioning and pointing determination system (PPDS) type (00 is "predict", 01 is "rapid", 02 and higher is "final"),
[release_number] = release number (002), representing the SOC SDS (software) release used to generate this L4A dataset,
[production_version] = granule production version , e.g., a particular data granule (or file) may have been regenerated multiple times,
[version_number] = L4A dataset production version (002), corresponding to the ORNL DAAC’s dataset version number, and
.h5 = file extension, HDF5 format.

Table 1. File names and descriptions.

File Name	Description
Data Files
GEDI04_A_2021188232338_O14550_04_T08520_02_002_01_V002.h5	Each contains information and data in METADATA and BEAMXXXX groups and three ANCILLARY group datasets.
Companion Files
GEDI_ATBD_L4A_v1.0.pdf	Algorithm Theoretical Basis Document (ATBD) for GEDI L4A Footprint Aboveground Biomass Density Product (current dataset).
GEDI_L4A_AGB_Density_V2.pdf	A PDF version of this user guide.
GEDI_L4A_V2_Common_Queries.pdf	Common data questions and answers on how to use and interpret the GEDI L4A product. This information is also provided in Section 5 of this user guide.
GEDI_L4A_V2_Product_Data_Dictionary.pdf	Data product dictionary that provides detailed information about each variable included in the data files.

File Organization

Each GEDI04_A granule contains information in METADATA and BEAMXXXX groups in addition to three compound datasets.

The METADATA group contains data set identification information.

The BEAMXXXX root group (Table 2) contains the AGBD prediction, associated uncertainty metrics, quality flags, and model inputs including the scaled and transformed GEDI02_A RH metrics and other information about the waveform for the selected algorithm setting group.

There is one BEAMXXXX group for each of the eight beams with valid data. The GEDI04_A Version 2 product uses GEDI02_A Version 2 as input, however, the algorithm setting group selection used for GEDI02_A Version 2 has been modified for Evergreen Broadleaf Trees in South America to reduce false-positive errors resulting from the selection of waveform modes above ground elevation as the lowest mode. The BEAMXXXX root group contains the AGBD prediction, associated uncertainty metrics, quality flags, the scaled and transformed GEDI02_A RH metrics, and other information about the waveform for the selected algorithm setting group.

• The BEAMXXXX / geolocation group (Table 3) contains elevation, latitude, longitude, and other information for each algorithm selection group (i.e., 1, 2, 3, 4, 5, 6, and 10).
• The BEAMXXXX / land_cover_data group (Table 4) The BEAMXXXX / land_cover_data group contains land cover data extracted from external data sources, including Landsat tree cover, Landsat water persistence, a modified version of MCD12Q1 V006 PFT, the world region identifier, the TanDEM-X global urban footprint classification, and leaf-off and leaf-on flags. The PFT and world region identifier used in L4A have been updated from the GEDI04_A Version 1 data.
• The BEAMXXXX / agbd_prediction group (Table 5) contains ancillary information, AGBD predictions in natural and transformed units, and associated prediction uncertainty for each algorithm setting group. Providing these data allows the user to evaluate and select alternative algorithm setting groups.

The ANCILLARY / model_data group (Table 6)  in the GEDI04_A data provides parameters and variables from the L4A models used to generate predictions. All the model parameters and uncertainty estimates (e.g. variance-covariance matrix of the model parameters) required as input to the GEDI04_B algorithm are also provided.

ANCILLARY / pft_lut (Table 7) and ANCILLARY / region_lut (Table 8) are look-up tables that link a numeric value from gridded PFT or world region to a descriptive text name.

What is the algorithm setting group?

Investigators applied a sophisticated post-processing algorithm to the received waveforms from the GEDI instrument to detect weaker waveform signals. The “algorithm setting group” defines the specific set of parameters used in an algorithm run. There are six (i.e., 1, 2, 3, 4, 5, 6, and 10^¥) defined groups.

Each algorithm run's output and externally-set parameters are available in the L2A data product, within the ‘rx_processing_a<n>’ subgroup. For details refer to Hofton and Blair (2020).

In the L4A data products, the geolocation group and the agbd_prediction group report footprint data for each “algorithm setting group”. The variables are *_aN, where N is 1, 2, 3, 4, 5, 6, or 10^¥. In the BEAMXXXX root group, the reported AGBD prediction value is for the selected algorithm setting group. The selected “algorithm setting group” is contained in the selected_algorithm variable. The selected AGBD value is reported in the agbd_prediction group dataset.

^¥ Note that a value of 10 indicates algorithm setting group 5 has been used, but that the lowest detected mode is likely a noise detection. When this occurs, a higher mode has been used to calculate RH metrics (Hofton and Blair, 2020).

Variables in the L4A Footprint Data Files

Data are inputs from L2A (Source L2A) and outputs of the GEDI04_A algorithm, descriptors, and quality flags. Data files are provided for each beam.

Table 2. Variable names and descriptions in the Aboveground Biomass Density group. These variables include the AGBD prediction, associated uncertainty metrics, quality flags, the scaled and transformed GEDI02_A RH metrics, and other information about the waveform for the selected algorithm setting group. Input variables from the GEDI02_A data product are marked with "L2A" as the source.

Variable	Units (Source)	Description
agbd	Mg/ha	Predicted aboveground biomass density (Mg/ha)
agbd_pi_lower	Mg/ha	Lower prediction interval (see alpha attribute for the level)
agbd_pi_upper	Mg/ha	Upper prediction interval (see alpha attribute for the level)
agbd_se	Mg/ha	Aboveground biomass density (Mg/ha) prediction standard error
agbd_t		Model prediction in fit units
agbd_t_se		Model prediction standard error in fit units (needed for calculation of custom prediction intervals)
algorithm_run_flag		The L4A algorithm is run if this flag is set to 1. This flag selects data that have sufficient waveform fidelity for AGBD estimation.
beam	(L2A)	Beam identifier
channel	(L2A)	Channel identifier
degrade_flag	(L2A)	Flag indicating degraded state of pointing and/or positioning information
delta_time	s (L2A)	Time since Jan 1 00:00 2018.
elev_lowestmode	m (L2A)	Elevation of center of lowest mode relative to reference ellipsoid
l2_quality_flag		Flag identifying the most useful L2 data for biomass predictions
l4_quality_flag		Flag simplifying selection of most useful biomass predictions
lat_lowestmode	degrees (L2A)	Latitude of center of lowest mode
lon_lowestmode	degrees (L2A)	Longitude of center of lowest mode
master_frac	s (L2A)	Master time, fractional part. master_int+master_frac is equivalent to /BEAMXXXX/delta_time.
master_int	s (L2A)	Master time, integer part. Seconds since master_time_epoch. master_int+master_frac is equivalent to /BEAMXXXX/delta_time.
predict_stratum		Prediction stratum identifier. Character ID of the prediction stratum name for the 1 km cell
predictor_limit_flag		Predictor value is outside the bounds of the training data (0=in bounds; 1=lower bound; 2=upper bound)
response_limit_flag		Prediction value is outside the bounds of the training data (0=in bounds; 1=lower bound; 2=upper bound)
selected_algorithm	(L2A)	Selected algorithm setting group
selected_mode	(L2A)	ID of mode selected as lowest non-noise mode
selected_mode_flag	(L2A)	Flag indicating status of selected_mode
sensitivity	(L2A)	Beam sensitivity. Maximum canopy cover that can be penetrated considering the SNR of the waveform
shot_number	(L2A)	Unique identifier used to link observations between groups and between data products. The shot number format is OOOOOBBRRGNNNNNNNN, where OOOOO is the orbit number, BB is the beam number, RR is reserved for the future and G is the sub-orbit number, and NNNNNNNN is the shot number within the beam
solar_elevation	degrees (L2A)	Solar elevation angle
surface_flag	(L2A)	Indicates elev_lowestmode is within 300m of Digital Elevation Model (DEM) or Mean Sea Surface (MSS) elevation
xvar		Predictor variables (offset and transformation have been applied)

Table 3. Variable names and descriptions in the Geolocation group. This group contains elevation, latitude, longitude, and other information for each algorithm selection group (i.e., 1, 2, 3, 4, 5, 6, and 10). Input variables from the GEDI02_A data product are marked with "L2A" as the source.

Variable	Units (Source)	Description
elev_lowestmode_aN	m (L2A)	Elevation of center of lowest mode relative to the reference ellipsoid.
lat_lowestmode_aN	degrees (L2A)	Latitude of center of lowest mode.
lon_lowestmode_aN	degrees (L2A)	Longitude of center of lowest mode.
sensitivity_aN	(L2A)	Maximum canopy cover that can be penetrated considering the SNR of the waveform.
shot_number	(L2A)	Shot number.
stale_return_flag	(L2A)	Flag from digitizer indicating the real-time pulse detection algorithm did not detect a return signal above its detection threshold within the entire 10 km search window. The pulse location of the previous shot was used to select the telemetered waveform.

Table 4. Variable names and descriptions in the Landcover group. This group contains land cover data extracted from external data sources, including Landsat tree cover, Landsat water persistence, a modified version of MCD12Q1 V006 PFT, the world region identifier, the TanDEM-X global urban footprint classification, and leaf-off and leaf-on flags. The PFT and world region identifier used in L4A are described further in Section 5 of this document. Input variables from the GEDI02_A data product are marked with "L2A" as the source.

Variable	Units (Source)	Description
landsat_treecover	percent (L2A)	Tree cover in the year 2010, defined as canopy closure for all vegetation taller than 5 m in height (Hansen et al., 2013) and encoded as a percentage per output grid cell.
landsat_water_persistence	percent	The percent UMD GLAD Landsat observations with classified surface water between 2018 and 2019. Values >80 usually represent permanent water while values <10 represent permanent land.
leaf_off_doy	days	GEDI 1 km EASE 2.0 grid leaf-off start day-of-year derived from the NPP VIIRS Global Land Surface Phenology Product.
leaf_off_flag		GEDI 1 km EASE 2.0 grid flag derived from leaf_off_doy, leaf_on_doy, and pft_class, indicating if the observation was recorded during leaf-off conditions in deciduous needleleaf or broadleaf forests and woodlands. 1=leaf-off, 0=leaf-on.
leaf_on_cycle		Flag that indicates the vegetation growing cycle for leaf-on observations. Values are 0=leaf-off conditions, 1=cycle 1, 2=cycle 2.
leaf_on_doy		GEDI 1 km EASE 2.0 grid leaf-on start day- of-year derived from the NPP VIIRS Global Land Surface Phenology product.
pft_class		GEDI 1 km EASE 2.0 grid Plant Functional Type (PFT) derived from the MODIS MCD12Q1v006 product. Values follow the Land Cover Type 5 Classification scheme.
region_class		GEDI 1 km EASE 2.0 grid world continental regions (0=Water, 1=Europe, 2=North Asia, 3=Australasia, 4=Africa, 5=South Asia, 6=South America, 7=North America).
shot_number	(L2A)	Shot number
urban_focal_window_size	pixels	The focal window size used to calculate urban_proportion. Values are 3 (3x3 pixel window size) or 5 (5x5 pixel window size).
urban_proportion	percent	The percentage proportion of land area within a focal area surrounding each shot that is urban land cover. Urban land cover was derived from the DLR 12 m resolution TanDEM-X Global Urban Footprint Product.

Table 5. Variable names and descriptions in the Aboveground Biomass Prediction group. This group contains ancillary information, AGBD predictions in natural and transformed units, and associated prediction uncertainty for each algorithm setting group (i.e., 1, 2, 3, 4, 5, 6, and 10). Providing these data allows the user to evaluate and select alternative algorithm setting groups.

Variable	Units	Description
pft_grid_version		1 km Plant Functional Type grid version
pft_infilled_grid_version		1 km Plant Functional Type prediction strata grid version
region_ grid_version		1 km geographic region prediction strata grid version
phenology_grid_version		1 km phenology metrics grid version
urban_grid_version		25 m urban proportion grid version
water_grid_version		25 m water persistence grid version
predictor_offset		Offset applied to predictors before model fitting
response_offset		Offset applied to the response before model fitting
l2a_alg_count		Number of L2A algorithm setting groups used for L4A
max_nvar		Maximum number of predictors in L4A models
alpha		Alpha value used for calculation of prediction intervals
agbd_aN	Mg/ha	Above ground biomass density; Geolocation latitude lowestmode (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_pi_lower_aN	Mg/ha	Above ground biomass density lower prediction interval (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_pi_upper_aN	Mg/ha	Above ground biomass density upper prediction interval (_aN=a1, 2, 3, 4, 5, 6, and a10).
predictor_limit_flag_aN		Predictor value is outside the bounds of the training data (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_se_aN	Mg/ha	Aboveground biomass density (Mg/ha) prediction standard error (_aN=a1, 2, 3, 4, 5, 6, and a10).
selected_mode_aN		ID of mode selected as lowest non-noise mode (_aN=a1, 2, 3, 4, 5, 6, and a10).
selected_mode_flag_aN		Flag indicating status of selected mode (_aN=a1, 2, 3, 4, 5, 6, and a10).
xvar_aN		Predictor variables (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_t_aN	Mg/ha	Aboveground biomass density model prediction in transform space (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_t_pi_lower_aN	Mg/ha	Lower prediction interval in transform space (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_t_pi_upper_aN	Mg/ha	Upper prediction interval in transform space (_aN=a1, 2, 3, 4, 5, 6, and a10).
agbd_t_se_aN		Model prediction standard error in fit units (_aN=a1, 2, 3, 4, 5, 6, and a10).
algorithm_run_flag_aN		Algorithm run flag-this algorithm is run if this flag is set to 1. This flag selects data that have sufficient waveform fidelity for AGBD estimation (_aN=a1, 2, 3, 4, 5, 6, and a10).
l2_quality_flag_aN		Flag identifying the most useful L2 data for biomass predictions (_aN=a1, 2, 3, 4, 5, 6, and a10).
l4_quality_flag_aN		Flag simplifying selection of most useful biomass predictions (_aN=a1, 2, 3, 4, 5, 6, and a10).
response_limit_flag_aN		Prediction value is outside the bounds of the training data (_aN=a1, 2, 3, 4, 5, 6, and a10).
shot_number		Unique identifier used to link observations between groups and between data products. The shot number format is OOOOOBBRRGNNNNNNNN, where OOOOO is the orbit number, BB is the beam number, RR is reserved for the future and G is the sub-orbit number, and NNNNNNNN is the shot number within the beam

Table 6. Variable names and descriptions in the Ancillary group: Model data. This group provides parameters and variables from the L4A models used to generate predictions. All the model parameters and uncertainty estimates (e.g. variance-covariance matrix of the model parameters) required as input to the GEDI04_B algorithm are also provided.

Variable	Units	Description
predict_stratum		Prediction stratum (e.g., DBT_Af=Deciduous Broadleaf Tree, Africa)
model_group		Model group (1= all predictors considered, 2 = no RH metrics below RH50, 3 = forced inclusion of RH98, 4 = forced inclusion of RH98 and no RH metrics below RH50)
model_name		Model name (prediction stratum used for the fit data)
model_id		Model rank used for the prediction stratum
bias_correction_name		Back-transform bias correction method (Snowdon, Baskerville)
bias_correction_value		Back-transform bias correction value
dof		Degrees of freedom
fit_stratum		Fit stratum
par		Model parameters (coefficients)
npar		Number of model parameters (coefficients)
predictor_id		Predictor identifier
predictor_max_value		Maximum value of predictor in transform space used to train the model
response_max_value	Mg/ha	Maximum value of Mg/ha used to train the model
rh_index		Index of RH metric to use as a predictor
rse		Residual Standard Error
vcov		Variance-covariance matrix of model parameters
x_transform		Predictor transform (sqrt, log, none)
y_transform		Response transform (sqrt, log)

Table 7. Variable names and descriptions in the Ancillary / pft_lut group. This group provides look-up tables that link a numeric value from gridded PFTs to a descriptive text name.

Variable	Description
pft_class	MCD12Q1 Type 5 plant functional type (PFT) class
pft_name	L4A Plant Functional Type strata

Table 8. Variable names and descriptions in the Ancillary / region_lut group. This group provides look-up tables that link a numeric value from the gridded world region to a descriptive text name.

Variable	Description
region_class	L4A geographical region identifier
region_name	L4A geographical region strata

Most previous efforts have developed site-specific or regional relationships between AGBD and remote sensing measurements (Drake et al., 2002). In contrast, GEDI requires models and algorithms designed to perform well throughout the entire observation domain of the ISS. Locally developed or regional relationships between AGBD and height are unlikely to perform well at locations outside the limited geographic extent of training data unless procedures are developed specifically to ensure transferability beyond the extent of calibration measurements. The GEDI L4A algorithm and product currently addresses two important components of transferability: (1) geographic transferability, meaning that the models can be extrapolated to locations outside the geographic extent of training data; and (2) transferability from simulated to recorded GEDI waveforms.

The GEDI Forest Structure and Biomass Database (FSBD) contained 31,414 simulated GEDI waveforms co-located with field plot estimates of AGBD. After excluding projects that are not analysis-ready or otherwise inappropriate for GEDI (e.g., variable radius plots), the unfiltered GEDI04_A calibration dataset contained 12,140 simulated GEDI waveforms. Quality control filters designed to flag observations that are likely to be erroneous (e.g., incongruence between height and AGBD) or that do not meet the requirements of the waveform simulator were then applied. The filtered GEDI04_A calibration dataset used to develop the second version of the GEDI04_A data product contained 8,587 simulated waveforms from 21 countries on all continents within the GEDI domain.

To quantify geographic transferability candidate models were evaluated within sets of 5-degree grid cells that contain simulated GEDI waveforms with coincident field data. Our approach sets aside data from one grid cell for testing and trains the model using data within the remaining grid cells. This model is used to predict AGBD within the held-out grid cell, and the process is repeated for all grid cells within each stratum for all models under consideration.

See the GEDI04_A ATBD (Kellner et al., 2021) for further details on the uncertainty/calibration analysis applied.

The GEDI instrument is aboard the International Space Station (ISS) and its mission aims to characterize ecosystem structure and dynamics to enable improved quantification and understanding of the Earth’s carbon cycle and biodiversity. GEDI is led by the University of Maryland in collaboration with NASA Goddard Space Flight Center. GEDI science data algorithms and products are created by the GEDI Science Team.

The GEDI instrument produces high-resolution laser ranging observations of the 3-dimensional structure of the Earth. GEDI was launched on December 5, 2018, and is attached to the ISS. GEDI collects data globally at the highest resolution and densest sampling of any light detection and ranging (lidar) instrument in orbit to date. The GEDI instrument consists of 3 lasers producing a total of 8 beam ground transects, which consist of ~25 m footprint samples spaced approximately every 60 m along-track. The GEDI beam transects are spaced approximately 600 m apart on the Earth’s surface in the cross-track direction, for an across-track width of ~4.2 km.

Footprint AGBD is derived from linear parametric models that relate GEDI L2A waveform relative height metrics to aboveground biomass estimates from co-located field plots. The GEDI approach to footprint model selection is data-driven. Candidate models are stratified by plant functional type (PFT) and continental region, with natural logarithm or square root transformations on the response and predictor variables. The GEDI footprint models represent the following combination of PFT’s—deciduous broadleaf trees, evergreen broadleaf trees, evergreen and deciduous needleleaf trees, and combinations of woodlands, grasslands, and shrubs.

GEDI footprint AGBD is an L4A data product (GEDI04_A). Models to produce GEDI04_A were developed using field estimates of AGBD colocated with simulated GEDI waveforms derived from discrete-return airborne lidar (Blair and Hofton, 1999; Hancock et al., 2019). The justification for using simulated GEDI waveforms is that few locations on the land surface are associated with field estimates of AGBD that could be used to train GEDI models. Because GEDI is a sampling mission and most field plots are small, GEDI data will not intersect most of these locations during the mission life. Simulated GEDI waveforms are processed to GEDI02_A equivalent relative height (RH) metrics, which are defined as the percentage of the received laser waveform intensity that is less than a given height, where height is computed relative to the elevation of the lowest mode in the waveform (Fig. 2).

Figure 2. Relative height (RH) metrics were calculated as the height relative to ground elevation under which a certain percentage of waveform energy has been returned. RH50, for example, is the height relative to the ground elevation below which 50% of waveform energy has been returned.

The GEDI approach to developing footprint AGBD models considers multiple candidates stratified by world region and PFT with different functional forms. The models were developed using a quality-filtered calibration dataset that contains 8,587 simulated waveforms in 21 countries. These data were contributed by numerous researchers and standardized into the GEDI FSBD, which is a living data archive that grows over time as new datasets are assimilated and improvements are made to existing records.

The GEDI04_A models are stratified by world region and PFT (Fig. 3). Important regions are under-represented in the GEDI FSBD, including the forests of continental Asia, the evergreen broadleaf forests throughout the islands of Southeast Asia and north of Australia, and the worldwide distribution of savannas and deciduous tropical forests.

Figure 3. The GEDI04_A global stratification of plant functional types (PFT) (A) and world region (B) used to produce GEDI footprint AGBD models. The box inset is the GEDI observation domain of 51.6 degrees N to S latitude. PFT: DBT (deciduous broadleaf trees), DNT (deciduous needleleaf trees), EBT (evergreen broadleaf trees), ENT (evergreen needleleaf trees), GSW (grasses, shrubs, and woodlands). Regions: Af (Africa), Au (Australia and Oceania), Eu (Europe), N-Am (North America north of southern Mexico), N-As (North Asia), S-Am (South America, Central America, and southern Mexico, and the Caribbean), S-As (South Asia).

GEDI04_A world region includes the geologically defined continents of Africa and Europe. The South America world region is the continent of South America, Central America and the Caribbean islands, and geological North America south of southern Mexico. The Australia and Oceania world region is geological Australia and the island regions north of Australia on the east side of the Wallace line, which defines the floral and faunal boundary between Australia and Asia during the Pleistocene (Mayr, 1944). The islands of Micronesia, Melanesia, and Polynesia are associated with the Australia and Oceania world region regardless of political affiliation. The North American world region includes geological North America north of southern Mexico. The continent of Asia was divided into north and south regions that approximately correspond to temperate and tropical forests.

GEDI04_A PFTs are assigned using an error-corrected and infilled 1 km grid derived from the Type 5 classification in the MODIS MCD12Q1 V006 data product (Friedl et al., 2002; 2010). These are deciduous broadleaf trees (DBT; class 4), deciduous needleleaf trees (DNT; class 3), evergreen broadleaf trees (EBT; class 2), evergreen needleleaf trees (ENT; class 1), and grasses, shrubs, and woodlands (GSW; classes 5 and 6).

On-orbit predictions of AGBD are made using the GEDI02_A elevation and height metric data product as input. The algorithms used by GEDI for generating these are described in the ATBD for GEDI Transmit and Receive Waveform Processing for L1 and L2 Products (Hofton and Blair, 2020). The L4A product contains the information necessary to reproduce the AGBD prediction for individual GEDI shots from L2A data; the algorithm setting group selection used in Version 2 GEDI02_A data is applied to these data on a per footprint basis.

The GEDI04_A algorithm is described in detail in the ATBD for L4 GEDI aboveground biomass density (Kellner et al., 2021), and the development of GEDI04_A models will be described in forthcoming publications. The algorithm generates a predicted value of AGBD in units of megagrams per hectare (Mg ha^-1) for every valid GEDI02_A waveform. The algorithm uses the latitude and longitude of the lowest mode to lookup the PFT from a modified version of MCD12Q1 V006 PFT classification and a world region grid. It then gets the selected estimator for the given combination of PFT and world region and predicts AGBD after scaling and transforming GEDI02_A RH metrics. Prediction intervals and the standard error of the prediction are generated and written to files.

Common Queries on How to Use and Interpret the GEDI04_A Data Product

This section is also provided in the companion file GEDI_L4A_V2_Common_Queries.pdf.

How are the GEDI04_A biomass estimates geolocated?

The GEDI04_A product uses the ground position as the location of each shot and AGBD estimate (elev_lowestmode, lat_lowestmode, lon_lowestmode). Additional waveform ranging points are available in the GEDI02_A product (e.g., elev_highestreturn, lat_highestreturn, lon_highestreturn) and may be joined to GEDI04_A using the shot_number dataset.

Note that the Version 2 GEDI04_A product is derived from the Version 2 GEDI02_A product (PGE Version 3), therefore has the same geolocation. Release 2 (PGE Version 3) GEDI02_A data product files are both available through the LP DAAC.

What quality metrics and flags should I use to filter the data?

AGBD is predicted for every shot where it is possible to run the GEDI04_A algorithm, as indicated by the algorithm_run_flag dataset (see Table 2). The GEDI04_A product provides multiple quality flags and metrics that may be used to subset the predictions to the most useful observations for a particular application or region.

The l2_quality flag encapsulates a number of GEDI02_A quality metrics to identify land surface shots with waveforms of high fidelity for AGBD estimation. The l4_quality_flag identifies shots that may be considered as samples of the population of which the applied models are representative. For example, GEDI04_A models for deciduous forests are only calibrated using GEDI waveforms simulated from leaf-on ALS data; therefore, we can only apply the derived models to on-orbit GEDI waveforms acquired under similar conditions.

The l2_quality flag uses a beam sensitivity threshold of 0.9 to match what is used for the Level 2 products. The l4_quality_flag uses a beam sensitivity threshold of 0.95, which was selected based on analysis of GEDI02_A and GEDI04_A on-orbit data. Beam sensitivity is an estimate of the maximum canopy cover that can be penetrated considering the signal-to-noise ratio of the waveform. For dense tropical forests, users may consider raising the beam sensitivity threshold (e.g., 0.98) to minimize measurement error in the RH metrics. In future versions, quality filtering will be improved by using the beam sensitivity together with the expected level of canopy cover for each shot.

Some users may wish to also evaluate the predictor_limit_flag and response_limit_flag. These identify shots with RH metrics or AGBD predictions, respectively, that are outside the observed range of values used to train the GEDI04_A models. Care should be taken when using such observations.

What are the units of xvar, and why doesn’t xvar match the relative height metrics in a corresponding GEDI02_A file?

The variables called xvar in the BEAMXXXX group and xvar_aN in the BEAMXXXX/agbd_prediction group are the scaled and transformed RH metrics used to generate the AGBD prediction for a given estimator and prediction stratum. GEDI_04A estimators are linear statistical models with a square root or natural logarithm transformation on the response or predictor variables. The appropriate transformation for the given estimator has been applied to GEDI_02A RH metrics to generate xvar and is indicated by the x_transform and y_transform variables in the ANCILLARY / model_data compound dataset. This transformation is applied after adding predictor_offset to the RH metrics. The predictor_offset is added because RH metrics can be negative when a large percentage of waveform energy is within the ground return. Because the square root and natural logarithm of a negative number are undefined, adding a large positive constant is necessary. For example, if a given estimator used a square root transformation, predictor_offset = 100, and the true RH metric had a value of 20, the number in xvar would be:

     xvar = √ (20+100)

What is the relationship between rh_index, predictor_id, and par in the ANCILLARY / model_data compound dataset?

The vector par contains coefficients of the linear model used to predict AGBD, where the first element is the intercept and subsequent elements are slope coefficients. The vector rh_index is the height percentile associated with the given RH metric. The variable predictor_id provides a mapping between rh_index and par. For example, if predictor_id is:

     predictor_ id = [1, 2, 3, 3, 0] 

and rh_index is:

     rh_ index = [50, 98, 50, 70, 0] 

the associated estimator (ignoring transformations) would be:

      AGBD = par[0] + par[1] x RH50 + par[2] x RH98 + par[3] x RH50 x RH70

Note that when the same predictor_id is associated with two rh_index values, it indicates that the product of two RH metrics was used in the given linear model. Note also that par[0] is always the intercept term.

How can I derive prediction intervals at a different confidence level?

The GEDI L4A product provides the standard error of the prediction and the lower/upper prediction intervals for every estimate. The default confidence level used for these intervals is 90%; however, some users may wish to specify their own confidence level. The general formula of a prediction interval for a new observation is:

     estimate ± (standard error x t-multiplier)

where the estimate is the sample prediction in transform space (agbd_t in Table 1) and standard error is the standard error of the prediction in transform space (see agbd_t_se in Table 1). The t-multiplier (t_1-α/2,dof) can be derived using standard libraries in R or Python and depends on: (1) the degrees-of-freedom for the applied model (dof), which is provided in the L4A product (ANCILLARY / model_data / dof); and (2) the t-distribution probability (α) which is specified by the BEAMXXXX / agbd_prediction group attribute alpha and may be modified by the user. For example, an alpha value of 0.1 is used for a 90% confidence level and 0.05 for a 95% confidence level.

Note that the prediction intervals described above are in transform space, and need to be back-transformed to place estimates in units of aboveground biomass density. A correction also needs to be applied to account for bias introduced by a transformation of the response variable (agbd). For example, if ANCILLARY / model_data/y_transform is “sqrt” and ANCILLARY / model_data/bias_correction_name is “Snowdon”, then

    agbd = agbd_t² * ANCILLARY/model_data/bias_correction_value

If ANCILLARY/model_data/y_transform is “exp” and ANCILLARY/model_data/bias_correction_name is “Baskerville”, then

     agbd=exp(agbd_t) * exp(ANCILLARY/model_data/bias_correction_value)

Blair, J.B., and M.A. Hofton. 1999. Modeling laser altimeter return waveforms over complex vegetation using high-resolution elevation data. Geophysical Research Letters 26:2509–2512. https://doi.org/10.1029/1999GL010484

Drake, J.B., R.O. Dubayah, R.G. Knox, D.B. Clark, and J.B. Blair. 2002. Sensitivity of large-footprint lidar to canopy structure and biomass in a neotropical rainforest. Remote Sensing of Environment 81:378–392. https://doi.org/10.1016/S0034-4257(02)00013-5

Dubayah, R., J.B. Blair, S. Goetz, L. Fatoyinbo, M. Hansen, S. Healey, M. Hofton, G. Hurtt, J. Kellner, S. Luthcke, J. Armston, H. Tang, L. Duncanson, S. Hancock, P. Jantz, S. Marselis, P.L. Patterson, W. Qi, and C. Silva. 2020. The Global Ecosystem Dynamics Investigation: High-resolution laser ranging of the Earth’s forests and topography. Science of Remote Sensing 1:100002. https://doi.org/10.1016/j.srs.2020.100002

Friedl, M.A., D.K. McIver, J.C.F. Hodges, X.Y. Zhang, D. Muchoney, A.H. Strahler, C.E. Woodcock, S. Gopal, A. Schneider, A. Cooper, A. Baccini, F. Gao, and C. Schaaf. 2002. Global land cover mapping from MODIS: algorithms and early results. Remote Sensing of Environment 83:287–302. https://doi.org/10.1016/S0034-4257(02)00078-0

Friedl, M.A., D. Sulla-Menashe, B. Tan, A. Schneider, N. Ramankutty, A. Sibley, and X. Huang. 2010. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sensing of Environment 114:168–182. https://doi.org/10.1016/j.rse.2009.08.016

Hancock, S., J. Armston, M. Hofton, X. Sun, H. Tang, L.I. Duncanson, J.R. Kellner, and R. Dubayah. 2019. The GEDI Simulator: A Large–Footprint Waveform Lidar Simulator for Calibration and Validation of Spaceborne Missions. Earth and Space Science 6:294–310. https://doi.org/10.1029/2018EA000506

Hansen, M.C., P.V. Potapov, R. Moore, M. Hancher, S.A. Turubanova, A. Tyukavina, D. Thau, S.V. Stehman, S.J. Goetz, T.R. Loveland, and A. Kommareddy. 2013. High-resolution global maps of 21st-century forest cover change. Science, 342(6160):850-853. https://doi.org/10.1126/science.1244693

Hofton, M.A., and J.B. Blair. 2020. Algorithm Theoretical Basis Document (ATBD) for GEDI Transmit and Receive Waveform Processing for L1 and L2 Products. Goddard Space Flight Center, Greenbelt, MD. https://doi.org/10.5067/DOC/GEDI/GEDI_WF_ATBD.001

Kellner, J.R., J. Armston, and L. Duncanson. 2021. Algorithm Theoretical Basis Document for GEDI footprint aboveground biomass density. See companion file GEDI_ATBD_L4A_v1.0.pdf.

Mayr, E., 1944. Wallace’s Line in the Light of Recent Zoogeographic Studies. The Quarterly Review of Biology 19:1–14. https://doi.org/10.1086/394684