Data Sources

Currently this website compares just two carbon maps, but we hope to expand this in the near future.

Baccini et al. (2012)

The Baccini et al. map was the first pantropical map of aboveground carbon produced at a 500 m resolution. It was published in the journal Nature Climate Change in 2012. A link to the original article can be found here.

The map is made from the following datasets:

• 283 field plots, located under GLAS lidar footprints
• 5.5 million canopy height estimates from the GLAS satellite
• MODIS BRDF data (vegetation colour)
• SRTM data (elevation)

The map was produced by combining these data using the algorithm RandomForest.

Saatchi et al. (2011)

The Saatchi et al. map is a map of aboveground carbon produced at a 1 km resolution, covering much of the tropics and subtropics. It was published in the journal Proceedings of the National Academy of Sciences in 2011 . A link to the original article can be found here.

The map is made from the following datasets:

• 493 calibration field plots
• 3 million canopy height estimates from the GLAS satellite (only used if at least 4 GLAS footprints fall within the same 1 km pixel)
• MODIS LAI/NDVI/VCT data (estimates of the leaf density, vegetation green-ness, and tree cover)
• SRTM data (elevation)
• QUIKSCAT data (seasonality and surface moisture)

The map was produced by combining these data using the algorithm Maxent.

Major differences in methodology

There are three major differences in methodology between the Baccini and Saatchi maps, which superficially at least use very similar input data (see above).

1. Allometric equations - both use allometric equations derived from the same source, a paper by Chave et al. (2005). However, Saatchi et al. use equations involving tree diameter, height and wood density, whereas Baccini et al. use equations invovling only diameter and height.
2. Treatment of GLAS data - both use scattered LiDAR footprints from the ICESat GLAS sensor as a key input dataset. Indeed, both maps could almost be considered a spatial extrapolation of the widely separated GLAS footprints. However, the two studies use the GLAS data very differently: Baccini et al. use a multiple regression to relate various characteristics of the lidar waveform to the biomass value of 283 field plots located under GLAS footprints; Saatchi et al. use 493 plots in the Amazon to relate the LiDAR waveform to Lorey's height (basal area weighted average height), and then use field plot data to convert these Lorey's height estimates into biomass using continent-specific equations.
3. Spatial modeling - The Saatchi et al. map uses Maxent to perform the spatial modeling and produce the final biomass estimates, whereas Baccini et al. use RandomForest.

These together must explain the large differences observed, particularly in central Amazonia, western Congo, southern Papua New Guinea, and savanna/woodland regions. It should be noted that these are regions where on average there are fewer ground calibration points, which not only explains part of the differences, but also means that verification is difficult.