MRV Carbon and Deforestation


Why Global Forest Watch is not suited for ensuring EUDR compliance

January 12, 2024

Written by

Caroline Busse

The Need for Data Integrity

The importance of applying reliable data in business-sensitive decisions, particularly in the context of due diligence and deforestation-free supply chains, is crucial. Accurate data is specifically needed for the upcoming implementation of the EU Regulation on Deforestation-Free Products (EUDR). From January 2025 onwards, companies must perform due diligence on the traceability of their commodity supply chains and ensure that their products do not stem from deforested land. Non-compliance with these requirements will lead to stringent penalties, including at least 4% of annual turnover fees.

The EUDR has caused a lot of turmoil, with companies progressing to implement their due diligence systems and service providers popping up to support them in checking the deforestation-free status of their plots.

At Nadar, we have noticed that the accuracy of the data applied in preparing for the upcoming regulation is often limited. One data product regularly used in this context is the public Global Forest Change data from Hansen et al. [1], which is available through Global Forest Watch (GFW).

Global Forest Watch Should not be Used for Compliance

Although GFW may be a good tool for conducting large-scale analyses of forest trends and patterns at the global, national, or regional level, it is essential to note that it should not be used for business-sensitive assessments as Hansen, Global Forest Watch (GFW) and scientific studies state [1],[2],[3].

We gathered some examples of satellite imagery to highlight the most striking issues in the GFW forest cover datasets.

No Differentiation between Natural Forests and Plantations

According to the EUDR, a 'forest' is defined as "land spanning more than 0,5 hectares with trees higher than 5 meters and a canopy cover of more than 10 %, [...] excluding land that is predominantly under agricultural or urban land use" and 'deforestation' is defined as "the conversion of forest to agricultural use, whether human-induced or not". [4]

Hansen and GFW explicitly warn against misunderstanding their tree cover definition. The Hansen data defines trees as "all vegetation taller than 5m in height." Its definition mixes natural forest and plantation tree crops as it includes "Oil palm, rubber, eucalyptus, and other managed stands qualify as tree cover if taller than 5 m." [3] This definition does not meet the definition of forest under the EUDR or other national forest definitions.

urban trees wrongly detected as forest cover in GFW data
The 2020 tropical tree cover layer in GFW includes urban trees in the city of Akim Oda in Ghana,

According to scientists, Hansen has "misclassified agricultural fields as forest [… ]" [5,6]. This misclassification has significant compliance implications, as EUDR-critical commodities such as oil palm, coffee, and rubber are among these crops.

The latest GFW dataset 'Tropical Tree Cover' from 2020 "does not disambiguate plantation trees from non-plantation trees". [7] Its definition mixes natural forest and plantation trees and thus cannot be used as a forest baseline layer for EUDR compliance. The EUDR mandates that agricultural lands should not be considered a part of the forest cover.

The 2020 tropical tree cover layer in GFW includes plantation land as tree cover in Southern Ghana,

Tree Cover Loss Does not Equal Deforestation

Many users assume that GFW data indicates deforestation when, in fact, it only shows tree cover loss. [3] GFW also advises its users to consider that none of the datasets available in GFW map deforestation as defined by the EUDR due to the inclusion of forest loss beyond agricultural conversion. It highlights the need for companies to carry out further due diligence. [8]

According to a CIFOR report, GFW data completely contradicts other data sources, such as the FAO's Global Forest Resource Assessment. In EUDR-relevant producing countries such as China, India, and Malaysia, the FAO attests to a comparably stable or increasing forest area, while GFW reports extensive tree cover loss. The significant losses in the GFW data can most likely be attributed to timber harvests. [9] These would not be marked 'deforestation' or 'degradation' under the EUDR if located on land designated for timber production before December 2020. GFW data in this context leads to a substantial overestimation of 'deforestation'.

The images below show a similar example from a smallholder coffee plantation in Peru, where GFW data indicates tree cover loss after 2020. The farm is classified as tree cover as Hansen does not differentiate between forest and agricultural land use. The tree cover loss likely stems from tree harvesting, pruning, and cutting of coffee trees.

The GFW 2020 tree cover data indicates forest loss on the plot after 2020 and misclassifies the coffee plantation as tree cover (in this case, we cannot share the coordinates due to client confidentiality).

Nadar's data clearly shows that the farm was already established before the cut-off date of December 2020 and is thus not considered 'deforestation' under the EUDR.

Nadar's EUDR Compliance platform shows the plot was farmland before the cut-off date
Nadar’s data shows that the plot was already farmland at the cut-off date of December 2020.

By relying on GFW data, this plot would be marked as deforested land and thus be non-compliant with the EUDR. Traders and importers would rely on false data and potentially exclude such a plot from their sourcing strategy to avoid risks.

Scientists agree that Hansen's data substantially underestimates actual deforestation as [it] "does not distinguish tropical forests from plantations". Several studies warn that results from Hansen "are misleading and can potentially lead to abuse", as "forest loss may be underestimated by tens of percents in the tropics" [6].

Underestimation of Deforestation on Smallholder Land

While Hansen's data may detect large-scale forest loss at global to regional scales, "it should be used cautiously because of the substantial commission error for small-scale disturbances [...]." [10] Hansen tends to underestimate forest loss, particularly in the case of small-scale disturbances of less than two hectares in size. [11] Therefore, companies cannot ensure compliance with the EUDR by using open data products like GFW, particularly when identifying deforestation on smallholder plots, such as in the case of cocoa, coffee, or rubber plantations where farmers, on average, manage a few hectares of land.

High Uncertainty in Sub-Saharan Africa

Studies show that Hansen's data vastly underestimates forest loss in Sub-Saharan Africa, where the accuracy is lowest, with 48 percent false negatives. This low accuracy is related to small-scale disturbances and the low-density canopy cover found in the subtropical biome, where it is harder to detect deforestation as the change from tree cover to bare ground is less noticeable than in a dense forest. [1],[2],[12] For traders and operators under the EUDR, this means that they lull themselves into a false sense of security as they assume that their plots are deforestation-free. Meanwhile, the actual deforestation rate might be much higher.

Multiple country-wide assessments show that GFW data significantly underestimates forest cover loss in Western Africa, such as in Côte d'Ivoire, where Hansen data indicates 11.600 hectares of primary forest loss in 2019 [13], against a report of Vivid Economics supported by the UK Space Agency [14] that states the country lost 68.000 hectares of primary forest in the same period; which constitutes a sixfold difference. Traders and operators would, in this case, assume that their plots are deforestation-free before being potentially caught up by the harsh reality of non-compliance through the EU Commission.

Overestimation of Deforestation in Tropical Ecosystems

An assessment by Forclime reports that Hansen "is hugely overestimating deforestation in Indonesia" due to a misclassification of forests. "Non-forest land such as smallholder agriculture, shrubland or plantations is wrongly classified as forest area". "Research based on the 'Hansen Map' has to be checked cautiously as it might produce wrong results [...]". In Indonesia, "Hansen detects high rates of phantom deforestation" resulting in an overestimation of deforestation by up to 347%. [15] In this case, traders and operators might wrongfully pull away from their suppliers as they assume that their plots contain deforestation and are thus EUDR non-compliant, while it was just an issue in the data.

Large Data Gaps in Tropical Tree Cover Data

he most recent '2020 Tropical Tree Cover' layer by Brandt et al. [7] available in GFW shows significant deficiencies, such as large data gaps covering up to dozens of kilometer-long patches of landscapes and even entire towns.

The data gaps are likely due to the cloud masking process of the new algorithm. These data gaps are specifically abundant in regions with high cloud cover, such as Sub-Saharan Africa, where most of the world’s cocoa is produced. These data gaps can be found in every country in Western Africa.

Large data gaps in the GFW 2020 tropical tree cover layer near Huni Valley in Southern Ghana,

Large data gaps in the GFW 2020 tropical tree cover layer near the city Enchi at the border of Western Ghana and Cote d'Ivoire,

Large data gaps in the GFW 2020 tropical tree cover layer near a large palm oil plantation in Ehania, Sud-Comoe, Southern Ivory Coast,

Similar issues exist in the main coffee-producing regions in Ethiopia, Brazil, and Vietnam and the primary palm oil and rubber-producing countries Malaysia, Indonesia, and Thailand.

Large data gaps in the GFW 2020 tropical tree cover layer near Tepi in Southern Ethiopia,

Data gaps in the GFW 2020 tropical tree cover layer in the coffee and cattle farming region encroaching the Bom Futuro National Forest in Rondônia, Brazil,

Should your farm plot lie in one of these data gaps, there will be no data available on the forest cover baseline of that plot, making GFW a limited and risky data source to rely upon.


Our assessment corresponds with the findings of scientists, and NGOs. While GFW datasets such as the Hansen data may be used to offer an initial understanding of global, national, or regional trends in forest cover or identify large-scale forest loss events in certain regions, it should not be the primary source for business-sensitive analyses, such as ensuring compliance with the EU Deforestation Regulation.

We advise you to be cautious of service providers without a background in remote sensing or forestry sciences offering deforestation monitoring and EUDR compliance, as they are most likely offering an interface to open data products such as GFW. What is needed is a reliable and thorough assessment to prove the deforestation-free status of the producing farms and plantations to ensure compliance with the EUDR.

[1] Hansen et al. (2013). High-Resolution Global Maps of 21st-Century Forest Cover Change. Science 342, 850-853.
[2] Global Forest Watch (2015): How accurate is accurate enough? Examining the GLAD global tree cover change data (Part 1, 2)
[3] Hansen et al. (2014). Response to Comment on “High-resolution global maps of 21st-century forest cover change”. Science 344, 981-981.
[4] Official Journal of the European Union (2023): Regulation (EU) 2023/1115 of the European Parliament and of the Council of 31 May 2023 on the making available on the Union market and the export from the Union of certain commodities and products associated with deforestation and forest degradation and repealing Regulation (EU) No 995/2010
[5] Cunningham et al. (2019). Identifying Biases in Global Tree Cover Products: A Case Study in Costa Rica. Forests 10, 10: 853.
[6] Tropek et al. (2014). Comment on “High-resolution global maps of 21st-century forest cover change”. Science 344, 981-981.
[7] Brandt et al. (2023). Wall-to-wall mapping of tree extent in the tropics with sentinel-1 and sentinel-2. Remote Sensing of Environment, 292, 113574.
[8] GFW (2023). 3 Ways Global Forest Watch Can Support the EU Law on Deforestation-free Supply Chains
[9] CIFOR, Peter Holmgren (2015). Can we trust country-level data from global forest assessments?
[10] Shimizu et al. (2020). Accuracy Assessments of Local and Global Forest Change Data to Estimate Annual Disturbances in Temperate Forests. Remote Sens. 12, 2438.
[11] Milodowski et al. (2017). Forest loss maps from regional satellite monitoring systematically underestimate deforestation in two rapidly changing parts of the Amazon. Environmental Research, 12, 094003.
[12] Tyukavina et al. (2015). Aboveground carbon loss in natural and managed tropical forests from 2000 to 2012. Environmental Research, 10, 074002.
[13] Global Forest Watch Dashboard Côte d'Ivoire
[14] UK Space Agency (2020), State and Trends of Deforestation in Côte d'Ivoire.
[15] FORCLIME (2017). The high-resolution global map of 21st-century forest cover change from the University of Maryland ('Hansen Map') is hugely overestimating deforestation in Indonesia.

Want to learn how our solution can support your business?

Learn more

Share this article

About the author

Caroline Busse


linkedin logo

Caroline is an experienced data scientist with a management degree from TU Munich and a degree in earth observation from the University of Würzburg, which is co-chaired by the German Aerospace Center (DLR). She has worked as a data scientist in the areas of nature conservation and land use change monitoring at WWF, the German Centre for Integrative Biodiversity Research (iDiv), and at tech companies such as Celonis and Deloitte.


Further Articles

Do you want to know more?

Connect with one of our experts to explore further or request a demo

Get a demo