Forest Carbon Diligence in the Amazon

By Christopher Anderson, Amy Rosenthal, and Flávia de Souza Mendes.

Today, Amazon Conservation released the first in a series of three Mapping the Andean Amazon Project (MAAP) reports on forest carbon across the Amazon that leverages our Forest Carbon Diligence product. The reports reveal patterns of aboveground carbon dynamics, including where and how much aboveground carbon has been lost - and gained - in the Amazon biome over the past 10 years.

Notably, this first report (MAAP report #215) estimates the biome's total aboveground forest carbon at about 56.8 billion metric tons, and suggests that Earth's most critical forest remains a carbon sink - though just barely. With the Amazon on the cusp, effective and timely action is urgently required to safeguard a globally important carbon sink and forestall a flip into a net carbon emissions source.

As will be expected by longtime readers, like we are, the findings both substantiate the best scientific understanding of trends and also identify some notable surprises, which we'll dive into below. Together, these conclusions indicate key opportunities for conservation, reason for hope, and actionable insights to forestall forest emissions.

You can find these reports on the MAAP website, where you can sign up to have future MAAP alerts delivered to your inbox. We're big fans of the data-driven environmental reporting produced by their team, and we highly recommend the read.

The Planet team was thrilled to work with Amazon Conservation on these summary reports. In this companion post, we aim to complement the reports with an explainer about how and why these insights came to light with the Forest Carbon Diligence datasets and to provide scientific guidance regarding how to evaluate the quality and veracity of these results.

First we'll share a few of our big takeaways from the first MAAP report (#215), followed by a series of FAQs raised by these findings:

1. As of 2022, total aboveground forest carbon in the Amazon biome can be estimated at about 56.8 billion metric tons. This is right in between the lower and upper estimates found in the scientific literature.
2. That's 64.7 million metric tons more than in 2013, making the Amazon a carbon sink over the last decade - not a net source of carbon emissions. That's a very small buffer, however, and there's reason to worry that the biome could flip from sink to source with ongoing deforestation.
3. High carbon density - or the amount of aboveground carbon per hectare - is found across the Amazon, but there are peak carbon levels in two geographically distant extremes: the southwestern Amazon including southeastern Peru and adjacent Brazil and eastern Guiana Shield including northeastern Brazil, French Guiana, and southeastern Suriname. This is notable, given how much higher the carbon density is, highlighting the outsized importance that targeted conservation could play.

From 2013 to 2022, we see anincrease in average carbon density in mature forests, particularly in the Guiana Shield countries of French Guiana, Suriname, and Guyana. This result challenges some conventional wisdom about 'saturation,' in which old growth forests are assumed to reach a point where carbon sequestration is balanced by carbon losses from respiration and decomposition. These results indicate that intact forests continue to accumulate carbon, increasing aboveground forest carbon over time, and that these effects are measurable.

What is Forest Carbon Diligence?

Planet's Forest Carbon Diligence products quantify-globally and annually-how much carbon is stored in trees, the area occupied by trees, and how tall they are. This is done using cutting-edge machine learning models that fuse historical satellite observations with high quality, laser-derived reference data. Model benchmarks include an extensive archive of high resolution airborne LiDAR data for the canopy height and canopy cover models, and a global carbon dataset derived from the Global Ecosystem Dynamics Investigation (GEDI), a NASA satellite mission. This product builds on decades of open data and open science, and was designed to maximize accuracy, transparency, and trust based on well-known standards.

You can think of Diligence as a multi-year, GEDI-like forest carbon data product with wall-to-wall spatial coverage. It includes historical time series data since 2013 with numerical estimates for aboveground carbon density, canopy height, and canopy cover at 30-meter nominal resolution. It provides pixel-level uncertainty data, QA data, and day-of-measurement data. You can read more about Diligence in the product's Technical Specifications.

How do Forest Carbon Diligence carbon density results relate to the peer-reviewed literature?

The first MAAP report estimates that the Amazon biome contains 56.8 billion metric tons of aboveground carbon. That's a big, mind-boggling number. Even bigger when we Americans convert it to pounds (lbs), which translates to over 125 trillion lbs of carbon. For perspective, that would be roughly equivalent to the biomass of 9 billion African elephants! And for additional perspective on that point, the total population of African elephants is only about 415,000, down from about 10 million in 1930, per a WWF report.

But those weren't the comparisons you were asking about. How does the Diligence estimate compare to other peer-reviewed estimates of carbon stocks?

The best source for globally evaluating how Diligence compares to other peer-reviewed datasets is the Diligence Validation and Intercomparison Report, which we described in a previous Planet Pulse post. But we'll need to look elsewhere for Amazon-specific comparisons. Here are a few we could track down, with the caveat that the boundary definitions for the Amazon biome (analyzed by MAAP) and the Amazon biome (analyzed by others) may not exactly match.

Gatti et al. (2023) recently published an excellent paper on how carbon emissions from land use change in the Amazon have shifted as a result of decreased law enforcement activities, particularly in the western Amazon. Their carbon data appears to cite Saatchi et al. (2007), who published carbon stock estimates for the year 2000. After adjusting some numbers-converting the biomass-to-carbon scalar from 0.5 to 0.48 to ensure consistency between datasets-Saatchi et al. estimated the total carbon stocks in the Amazon basin to be between 56.0 and 69.3 billion metric tons (Pg). Diligence estimates 56.8 Pg for the year 2022, which follows around 20 years of deforestation in the region. This seems to be a reasonable level of agreement.Ometto et al. (2023) separately developed estimates of carbon density across the Brazilian Amazon, derived from fusing airborne LiDAR, field plots, and satellite data. They estimate average aboveground biomass density in the Amazon basin to be 174 metric tons per hectare (Mg/ha) or, converted to carbon density, 83.5 Mg/ha. Diligence estimates are lower than Ometto et al. Why is there misalignment? Both methods estimate carbon from forest structure data, but each does so with a different approach. Diligence uses GEDI as a reference dataset, while Ometto et al. uses regional allometry. Regionally calibrated estimates usually provide finer, local-scale calibration, but the field-to-LiDAR fits in this case appear to be very noisy, and we can't speak to the level of agreement between these results and other independent datasets.

^{Figure 1. Comparisons of FAO, Diligence, and GEDI L4B estimates of aboveground biomass density across Brazil. Please note that the numbers reported in these tables and figures are in units of aboveground biomass and not aboveground carbon, which requires a scale conversion for consistency with the MAAP reports.}

What about other sources that could provide a standard reference, like National Forest Inventory data? Figure 1, sourced from a prior analysis, compares aboveground biomass density and total biomass estimates to the Food and Agriculture Organization of the United Nations (FAO) and GEDI across all of Brazil. We found strong agreement between Diligence and GEDI across Brazil, including an r-squared score of 0.84. Diligence estimates are right between the national-scale estimates from FAO and GEDI. Analytically, we typically feel good about estimates that land between multiple independent results, which is the case here. More country-level GEDI/FAO comparisons are provided in Section 8 of the validation report, and Diligence does not appear to be systematically higher or lower than either source.

How confident are we in the finding that the Amazon biome remains a carbon sink?

Determining whether the Amazon is a net source or a net sink of CO₂ has been an active research topic for the past few years. Making the transition from sink to source could mark a critical tipping point, where feedback loops between deforestation, drought, and fire would lead to progressively slower rates of carbon sequestration and evapotranspiration, and eventually towards large-scale turnover in species communities (Lovejoy and Nobre, 2018). This is often referred to as the 'savannization' of the Amazon (Silvério et al. 2013).

So the finding from MAAP that the Amazon remains a net carbon sink is reason for cautious optimism, even while forest loss in the region has remained high over the past decade (World Resources Institute, 2024). Are these results consistent with other studies?

Two important papers provide evidence on both sides of the debate. Baccini et al. (2017) published one of the first studies to provide evidence via remote sensing that the tropics had transitioned to become a net carbon source. They modeled carbon stocks using 500-meter (m) resolution MODIS data, then fit time series trends to identify whether a pixel experienced loss, gain, or no change. The next paper is from Harris et al. (2021), who reported that both Brazil and the South American tropics overall remain a net carbon sink, using 30m Landsat data. Table 1 and Figure 2 from the Harris et al. paper are worth your attention.

What's the difference between these papers? It's mostly in how carbon gain gets estimated.

• Harris et al. used process-based carbon sequestration models, which are not satellite-based, but should give good estimates on average.
• Baccini et al. used a statistical smoothing process to denoise the signal over time, which may minimize small changes in forest growth while being more sensitive to forest loss.

We won't pick a side between the two, which are both excellent, innovative contributions to the field. But our results represent another contribution to the discussion, which lands pretty much right in between the papers cited above. MAAP found the Amazon is a net sink-by a small margin of just +64 million metric tons. Relative to the total estimate of 56.8 billion metric tons (Pg), net positive change is around +0.1%. To end up between the extremes of two sides of the debate seems a reasonable place to land.

What makes the Forest Carbon Diligence approach different is that no assumptions are made about growth rates, minimal time series smoothing is applied, and the maps are derived from multiple 30m resolution satellite datasets. Also, the shifts in carbon storage are directly attributed to estimated changes in canopy height and canopy cover. This is opposed to trying to estimate carbon stocks directly from optical data, which can be troublesome (Ploton et al. 2020).

We interpret these results to suggest that, while deforestation and forest degradation are major drivers of carbon emissions across the Amazon, it appears that-in addition to gains from tree cover expansion-increases in carbon accumulation in low-disturbance landscapes might offset these effects, which can be detected via small increases in canopy height and canopy cover.

With the Amazon approaching a tipping point, it's not clear how much longer we can expect the benefits of these carbon sink dynamics to last, especially with total forested area shrinking over time. These patterns reinforce the critical importance of conservation and stewardship in the region's remaining forests, especially the recognition and protection of land tenure rights in Indigenous lands (Prioli Duarte et al. 2023).

Figure 2. Ten-year time series of aboveground carbon density over Amazonas, Brazil showing trends in forest loss in response to agricultural expansion. Year-to-year variation in undisturbed landscapes also observed.

Is carbon density increasing in the Amazon, as Forest Carbon Diligence estimates?

Interpreting aggregate carbon density numbers is a tricky endeavor, so let's break down what they mean. Decreases in carbon density are primarily driven by forest loss via deforestation, selective logging, or natural disturbance. Increases in carbon density are observed where forests regrow or where existing forests continue to sequester and store carbon, which may be expressed through increases in canopy cover and canopy height.

The MAAP report quantifies shifts in average carbon density across the countries in the Amazon basin. Whether carbon density increases or decreases on average is a reflection of the balance between the drivers of loss and the drivers of gain. One remarkable trend here is the increase in carbon density in French Guiana and Suriname, with the highest and second-highest average carbon density in the Amazon, respectively. These countries are designated as high forest, low deforestation (HFLD), indicating the majority of forested areas are intact, mature forests. Potapov et al. (2017) and Global Forest Watch characterize much of these countries' forest regions as Intact Forest Landscapes. Minimizing deforestation in these regions could have a disproportionately positive impact on increasing carbon stocks. Is there corroborating evidence showing similar increases in carbon density across the region? Do intact forests continue to sequester and store carbon, or have they 'saturated' and reached a steady-state equilibrium?

Pan et al. (2024) provides the most contemporary evidence of net increases in carbon density in mature forests, in both intact tropical forests and in regrowing tropical forests, which builds on earlier work by Pan et al. (2011). The rates of carbon uptake in tropical forests are decreasing relative to sequestration rates in the 1990s when intact forests made up a larger portion of total forest area, but remain net positive overall.

What about evidence from the field? Duque et al. (2023) analyzed trends in the intact montane forests of the Andes, finding strong, persistent net increases in carbon density over an 11-year period. The authors suggest that "[their] results indicate that the Andes are similar to other tropical forests in that they are acting as aboveground carbon sinks, but the overall relative strength of the Andean carbon sink (1.01% annually) is even stronger than that of mature lowland tropical forests in Amazonia, Africa, or southeast Asia."

Based on long-term field plots from the RAINFOR network, Phillips and Brienen (2017) found that intact forests in the Amazon have sequestered enough carbon to offset both land-based emissions and national fossil fuel emissions. The synthesis provided in this work is truly remarkable, summarizing decades of plot measurements from hundreds of sites throughout the region:

The sink of carbon into mature forests has been remarkably geographically ubiquitous across Amazonia, being substantial and persistent in each of the five biogeographic regions within Amazonia. Between 1980 and 2010, it has more than mitigated the fossil fuel emissions of every single national economy… While the sink has weakened in some regions since 2000, our analysis suggests that Amazon nations which are able to conserve large areas of natural and semi-natural landscape still contribute globally-significant carbon sequestration.

These studies are all consistent with the results reported by MAAP. Planet developed the Forest Carbon Diligence product to provide these sorts of detailed, broad-scale insights into a dynamic carbon balance at local, regional, and global levels, anywhere in the world. But such a product requires a strong linkage to fundamental ecological research, like the field studies above. Satellite-derived products should be analyzed as a complement to, not a replacement for, such work.

How do we characterize uncertainty around Forest Carbon Diligence results in the Amazon?

Diligence provides pixel-level uncertainty quantification, estimated independently for canopy height, canopy cover, and aboveground carbon density. Users can draw on these pixel-level uncertainties to derive their own project-level estimates. Pre-computed estimates of total carbon uncertainty at project, biome, or country levels are not currently provided.

Users will need to make certain assumptions to develop their own aggregate uncertainty estimates. The crux is that one must know the spatial covariance among pixels if aggregating within a year, and potentially the spatiotemporal covariance if aggregating across both pixels and years. The covariance parameters may also be uncertain. Estimating spatial covariance often requires specifying a parametric model, which necessitates domain expertise specific to regions and forest types, and is computationally intensive. There are many different approaches to this kind of estimation, and the Forest Carbon Diligence products provide the raw materials required-robust, well-calibrated, distribution-free pixel-level uncertainties-for users to make decisions that are appropriate for their needs.

Fortunately, it is much more straightforward to estimate the total carbon over an area, which can be computed simply by summing the expected values for each pixel and then adjusting for area to get pixel-level MgC values.

To get a sense for the relative amount of uncertainty between Diligence and other datasets, we recommend reading more in Section 2 of the validation report. This section shows spatially-explicit maps of agreement among data sources, showing particularly strong agreement between Diligence and GEDI in the Amazon.

Where does training data come from for Forest Carbon Diligence, and is there any from the Amazon?

^{Figure 3. Extent and count of the airborne LiDAR data used for model training and evaluation. Source: Forest Carbon Diligence Technical Specifications.}

Diligence models were trained using publicly available data sourced over the last decade. The Planet team has developed an unparalleled data catalog of global, high resolution airborne LiDAR data, which was used for model training. This included hundreds of individual LiDAR sites across the Amazon basin and the Neotropics. Airborne LiDAR data were used to model canopy height and canopy cover. To estimate carbon density, we leveraged GEDI, a spaceborne LiDAR mission, which provides more than 5 years of large-footprint observations across the region. As a result, we observe some of the strongest agreement between Diligence and other independent data sources in the region.

These MAAP reports, drawing on Forest Carbon Diligence, explore the nuanced dynamics of forest carbon in the Amazon. There's no single story or trend that encapsulates the complexity and dynamism of this system. But with time series tracking of forest canopy and forest cover, we can get closer to actionable insights across the dominant trends that influence biome's health overall - including its provision of crucial ecosystem services for local communities, the countries they inhabit, and everyone who lives under our shared sky.

For forest carbon in the Amazon, not all locations are equal. These MAAP analyses show that mature forests continue to accumulate carbon over time. This points to the importance of safeguarding these biodiversity-rich, old growth forests to preserve the Amazon's carbon sink, alongside efforts to avoid further frontier forest losses and reforest what's been lost. The stability of Earth will be determined by our ability to understand its changing surface and intervene where necessary. MAAP report #215 reveals cutting-edge and actionable information to underpin data-driven conservation in the Amazon. Its overall message is simple: there's no time to lose.

Forward-looking Statements

Except for the historical information contained herein, the matters set forth in this blog post (including statements related to the Company in the third party blog post reproduced below) are forward-looking statements within the meaning of the "safe harbor" provisions of the Private Securities Litigation Reform Act of 1995, including, but not limited to, the Company's ability to successfully design, build, launch and deploy, operate and market new products and satellites and the Company's ability to realize any of the potential benefits from product and satellite launches, either as designed, within the expected time frame, in a cost-effective manner, or at all. Forward-looking statements are based on the Company's management's beliefs, as well as assumptions made by, and information currently available to them. Because such statements are based on expectations as to future events and results and are not statements of fact, actual results may differ materially from those projected. Factors which may cause actual results to differ materially from current expectations include, but are not limited to: the Company's ability to obtain and maintain required licenses and approvals from regulatory agencies, such as the Federal Communications Commission (FCC), in a timely fashion, or at all; whether the Company will be able to successfully build, launch and deploy or operate its satellites, including new satellites either as designed, in a timely fashion or at all; the Company's ability to develop and release product and service enhancements to respond to rapid technological change, or to develop new designs and technologies for its satellites, in a timely and cost-effective manner; whether the Company will be able to continue to invest in scaling its sales organization, expanding its software engineering (including its ability to integrate new satellite capabilities) and marketing capabilities; whether the Company will be able to accurately predict and capture market opportunity; whether current customers or prospective customers adopt the Company's platform or new products; the Company's ability realize any of the potential benefits from new products and satellites, as well as strategic partnerships and customer collaborations; and the risk factors and other disclosures about the Company and its business included in the Company's periodic reports, proxy statements, and other disclosure materials filed from time to time with the Securities and Exchange Commission (SEC) which are available online at www.sec.gov, and on the Company's website at www.planet.com. All forward-looking statements reflect the Company's beliefs and assumptions only as of the date such statements are made. The Company undertakes no obligation to update forward-looking statements to reflect future events or circumstances.

Public link

Please use the above public link if you want to share this noodl on another website.