Initial conditions for the ED2 model, derived from airborne laser scanning, Brazilian Amazon, 2016

Summary

This dataset provides regional distribution forest structure characteristics across the Brazilian Amazon that were derived from 545 airborne lidar transects (300 x 12500 m each) acquired during the Amazon Biomass Estimation Project (EBA2016) campaign in 2016, and described in Görgens et al. (2021). These datasets contain vertical distributions of stem number density, aboveground biomass and potential leaf area index for over 1,300,000 columns (50 × 50 m each) that were aggregated into 288 grid cells (1×1°) based on the geographic location. The data are reported in text format that is fully compatible with the initial conditions files for initialising the Ecosystem Demography Model (ED2; Longo et al. 2019), using the method described in Longo et al. (2020).

You need to download and uncompress all three files to use it with the ED2 model.

Relevant publication

Longo, M., M. Keller, L. M. Kueppers, K. Bowman, O. Csillik, A. Ferraz, P. R. Moorcroft, J. P. Ometto, B. S. Soares-Filho, X. Xu, M. L. F. de Assis, E. B. Görgens, E. J. L. Larson, J. F. Needham, E. M. Ordway, F. R. S. Pereira, E. Rangel Pinagé, L. Sato, L. Xu, and S. Saatchi. 2025. Degradation and Deforestation Increase the Sensitivity of the Amazon Forest to Climate Extremes, in review.

Dataset Overview

This dataset provides regional distribution forest structure characteristics across the Brazilian Amazon that were derived from 545 airborne lidar transects (300 × 12500 m each) acquired during the Amazon Biomass Estimation Project (EBA2016) campaign in 2016. These datasets contain vertical distributions of stem number density, aboveground biomass and potential leaf area index for over 1,300,000 columns (50 × 50 m each) that were aggregated into 288 grid cells (1 × 1°) based on the geographic location. The data are reported in text format that is fully compatible with the initial conditions files for initializing the Ecosystem Demography Model (ED2).

To generate this dataset, the discrete-return point cloud of each airborne lidar transect was split into multiple 50 × 50 m columns. Following the same methodology as in Longo et al. (2020), we used a waveform simulator and a light extinction model to retrieve vertical profiles of plant area density (PAD), and used the characteristics of the PAD curves to split the profiles into multiple size-dependent cohorts. Based on the cohort height and plant area index (PAI), we estimated leaf area index, stem number density and biomass of each cohort (Longo et al., 2020). We assigned each column to one grid cell (1 × 1°) based on the geographic location of each column. This resulted in 288 grid cells with at least one vertical profile across the Brazilian Amazon. The vertical profiles were also assigned to different sites within the grid cell, which are based on soil characteristics, as described in Longo et al. (2025).

Data characteristics

Spatial Coverage: Brazilian Amazon Biome (74°W–45°W; 14°S–5°N)

Spatial Resolution: 1° (grid cell) and sub-grid samples (50 m each)

Temporal coverage: 2016

Temporal Resolution: N/A

Data file information

There are 288 sets of files organized in 3 TAR/GZIP archives corresponding to the level of aggregation (sites, patches, cohorts). Each grid cell set contains 3 ASCII files, with extensions sss (site file), pss (patch file), and css (cohort file). These three files are formatted following the Ecosystem Demography Model convention for multi-site initial conditions (NL%IED_INIT_MODE=8) and described in detail here.

The files are named amzbr_GGGG.latLATlonLON.EEE, where

GGGG = grid cell identifier

LAT = latitude of the grid cell centre in decimal degrees

LON = longitude of the grid cell centre in decimal degrees

EEE = file extension (either sss, pss, or css)

For example, the file amzbr_0006.lat-12.5lon-60.5.css contains the cohort-level forest structure for grid cell ID 6, located at 12.5°S and 60.5°W. Note that not every ID between 1 and 350 exists, because some grid cells had no overlap with any lidar transect. The contents of each patch, site, and cohort file are described in Tables 1–3.

Data file details

Table 1. Variables in the site files (extension sss)

Variable

Unit

Description

time

years

Year

site

string

Unique site identifier, consistent with pss/css files for the same grid cell

area

fraction

Fraction of the grid cell area represented by site

depth

m

Soil depth to bedrock (Pelletier et al., 2016)

nscol

integer

Soil colour (Lawrence et al., 2007). Soil colour classes describe the albedo. Possible values go from 1 to 20, with 1 being the class with the highest albedo, and 20 being the one with the lowest albedo.

ntext

integer

Soil texture class (Poggio et al., 2021). Possible values, based on ED2 defaults (Longo et al., 2019) are

1. Sand

2. Loamy sand

3. Sandy loam

4. Silt loam

5. Loam

6. Sandy clay loam

7. Silty clay loam

8. Clay loam

9. Sandy clay

10. Silty clay

11. Clay

14. Silt

15. Heavy clay

16. Clayey sand

17. Clayey silt

sand

fraction

Sand fraction (Poggio et al., 2021)

clay

fraction

Clay fraction (Poggio et al., 2021)

slsoc

kg kg−1

Mass fraction of soil organic carbon (Poggio et al., 2021)

slph

pH

Soil acidity (Poggio et al., 2021)

slcec

mol kg−1

Cation exchange capacity (Poggio et al., 2021)

sldbd

kg m−3

Dry bulk density (Poggio et al., 2021)

elevation

m

Terrain elevation (dummy value, reserved field for future ED2-TOPMODEL implementation)

slope

degrees

Terrain slope  (dummy value, reserved field for future ED2-TOPMODEL implementation)

aspect

degrees

Terrain aspect (dummy value, reserved field for future ED2-TOPMODEL implementation)

TCI

dimensionless

Topography convergence index (dummy value, reserved field for future ED2-TOPMODEL implementation)

moist_f

m−1

Rate of exponential decay of soil conductance with depth (dummy value, reserved field for future ED2-TOPMODEL implementation)

moist_w

dimensionless

Soil wetness index (dummy value, reserved field for future ED2-TOPMODEL implementation)

Table 2. Variables in the patch files (extension pss).

Variable

Unit

Description

time

years

Year

site

string

Unique site identifier (consistent with sss/css files)

patch

string

Unique patch identifier (consistent with sss/css files). Patch names use the following convention: EBA_TNtttt_yyyy_uuu_Xeeeeeeee_Ynnnnnnnn where

tttt = EBA2016 transect number

yyyy = year of the lidar acquisition

uuu = UTM zone (and hemisphere)

eeeeeeee = Easting of the column centre (m)nnnnnnnn = Northing of the column centre (m)

dtype

integer

Disturbance type. Allowed classes are 

1. Pasture

2. Forest plantation

3. Tree fall

4. Burnt patch

5. Abandoned managed land

6. Logged forest (felling)

7. Skid trails (felling)

8. Cropland 

age

years

Patch age since last disturbance

area

fraction

Fractional area represented by patch. The sum of the areas of all patches within the same site add up to 1.

fgc

kgC m−2

Fast soil carbon (above ground). Estimated from land use history and literature search (Longo et al., 2020).

fsc

kgC m−2

Fast soil carbon (below ground). Estimated from land use history and literature search (Longo et al., 2020).

stgc

kgC m−2

Structural soil carbon (above ground). Estimated from land use history and literature search (Longo et al., 2020).

stgl

kgL m−2

Structural soil lignin (above ground). Estimated from land use history and literature search (Longo et al., 2020).

stsc

kgC m−2

Structural soil carbon (below ground). Estimated from land use history and literature search (Longo et al., 2020).

stsl

kgL m−2

Structural soil lignin (below ground). Estimated from land use history and literature search (Longo et al., 2020).

msc

kgC m−2

Microbial soil carbon. Estimated from land use history and literature search (Longo et al., 2020).

ssc

kgC m−2

Humified (slow) soil carbon. Estimated from land use history and literature search (Longo et al., 2020).

psc

kgC m−2

Passive (very slow) soil carbon. Estimated from land use history and literature search (Longo et al., 2020).

fsn

kgN m−2

Fast soil nitrogen (below and above ground). Estimated from land use history and literature search (Longo et al., 2020).

msn

kgN m−2

Mineralised soil nitrogen. Estimated from land use history and literature search (Longo et al., 2020).

npl

m−2

Total stem number density in this patch (consistent with the css file).

agb

kgC m−2

Aboveground biomass carbon density in this patch (consistent with the css file).

lai

m2 m−2

Leaf area index in this patch (consistent with the css file)

Table 3. Variables in the cohort files (extension css).

Variable

Unit

Description

time

years

Year

site

string

Unique site identifier (consistent with sss/pss files)

patch

string

Unique patch identifier (consistent with sss/pss files).

cohort

Integer

Cohort index in the patch

dbh

cm

Diameter at breast height

height

m

Plant height

pft

integer

Plant functional type, consistent with ED2 default classes. Possible values are:

1. Tropical C4 grass

2. Early-successional, evergreen broadleaf tropical tree

3. Mid-successional, evergreen broadleaf tropical tree

4. Late-successional, evergreen broadleaf tropical tree

nplant

m−2

Total stem number density of this cohort

bdead

kgC

Biomass stored in heartwood (structural, dead tissues)

balive

kgC

Biomass stored in living tissues (leaves, fine roots, sapwood, bark)

agb

kgC

Aboveground biomass of this cohort

lai

m2 m−2

(Maximum) Leaf area index of this cohort

Application and derivation

These files are intended to provide initial conditions for cohort-based vegetation demography models. Vegetation structure data (stem number density, aboveground biomass, leaf area index) were obtained from airborne lidar surveys following the method described in Longo et al. (2020) and extensively cross-validated. Soil edaphic characteristics are obtained from existing datasets (Lawrence et al., 2007, Pelletier et al., 2016; Poggio et al., 2021), and users are referred to these publications for uncertainty. Carbon stored in litter and soil layers (patch files) were estimated from the land use history and limited measurements in different land use types. These data are provided here because these quantities are required by ED2, but their uncertainty is likely very high and unlikely to provide reliable information at regional scale.

Quality assessment

Results using regional cross-validation (i.e., bootstrapping sampling that sets entire regions as testing) showed that the approach produces realistic variability in forest structures in the Amazon, both across precipitation gradients and at different levels of forest degradation (Longo et al., 2020).

Data Acquisition, Materials, and Methods

To turn raw airborne lidar dataset into forest structure distributions, we used the same approach described in Longo et al. (2020). We pooled airborne lidar data from the EBA campaign archive, publicly available through Ometto et al. (2023), selected only the transects designated for random sampling (545 transects). Each transect was split into multiple 50 × 50 m columns, resulting in 1,310,478 columns. Each column was assumed to be one patch. For each patch, we simulated waveforms using an approach similar to the GEDI waveform simulator (Hancock et al., 2019) and applied a light extinction model based on Ni-Meister et al. (2001) to derive a first guess of leaf area density (LAD). We used the lidar-based statistical models derived from airborne lidar metrics (Longo et al., 2020) to estimate aboveground biomass carbon density (ABCD), basal area (BA), leaf area index (LAI), and stem number density (ND). The first-guess LAD profiles were assigned a correction factor to minimise the overall uncertainty of ABCD, BA, LAI and ND (Longo et al., 2020).

Because the airborne lidar campaign did not survey deforested areas (Ometto et al., 2023), we complemented the lidar profiles with additional patches that accounted for deforested areas. For simplicity, we assumed that these areas were pastures entirely covered with C4 grasses, and further assumed an LAI of 2m2 m−2, based on values typically found in the Amazon (Zanchi et al., 2009).

To account for the associations between forest structure and soil properties, we assigned each patch to a set of soil depth (derived from Pelletier et al., 2016), soil colour (Lawrence et al., 2007) and other soil characteristics from SoilGrids250m version 2.0 (Poggio et al., 2021). For each grid cell, we defined up to four sets of soil characteristics, which became sites, and assigned patches to each of the sites based on the similarity of soil characteristics.