About PalmWatch

Consumer brands release disclosures of the palm oil mills in their supply chain. These disclosures typically list data such as the mill name, parent company and group, location (country, latitude and longitude points, etc.), and Universal Mill List (UML) identifier. Brands publish disclosures once or twice a year, usually as a PDF document. In their original form, these documents are not machine readable, and we use a data pipeline to transform the PDF into a data table. Each brands' disclosures are available for download in full on their brand page.

Each mill from brand-published disclosures is linked to the industry-standard Universal Mill List (UML) dataset. This dataset includes various data about the mills, including geospatial coordinates (latitude and longitude), country and province, mill name, parent company, and company group. To learn more about this dataset, see Global Forest Watch's data page.

• Mills: The local facilities that crush fruit to produce crude palm oil. Search below for specific mills and to learn more about their production and deforestation impact.
• Mill Owners: The companies that directly control crushing mills that supply palm oil to consumer brands. Search below to learn more about their operations and the brands they supply.
• Mill Groups: The parent companies that control crushing mills that supply palm oil to consumer brands. Search below to learn more about the mills they control and the brands they supply.

To estimate the area of forest that a mill might have access to, we use a standard methodology for all mills across the world. We start by looking at the road network data (OpenStreetMap) in a 50 kilometer radius around the mill. 50km is a standard distance in palm oil supply chain analysis that approximates the distance palm oil plantation agricultural products can travel to a processing mill.

For each mill, we find all the parts of the road network closest to that mill. In some cases, the road network may only be with 50km of one mill; otherwise, which mill is closest will be assigned that section of the road network. Based on those parts of the road network, we then create a single shape (polygon) generated from Voronoi tessellation of the assigned network points. At this step, each mill has a shape around it that represents its access to the road network and the estimated plantation areas.

Once each mill has a polygon catchment area, we use that shape to extract data from Global Forest Change dataset published by Hansen, Potapov, Moore, Hancher et al at the University of Maryland. This data provides estimates tree cover loss by year (2000 to 2022) and total forest cover levels as of 2000. We transform this geospatial raster data into summary statistics, which represent the estimated forest loss over time in the area assigned to that mill.

We calculate three risk scores based on the tree cover loss in each mill's assigned catchment area. Each risk score takes z score of the relevant value and standardizes that to a score of 1 to 5, where 1 represents low risk, and 5 represents high risk.

• Recent Deforestation Score: The current risk score reflects tree cover loss during the two most recent years of data (2021 and 2022).
  1. Calculate average tree loss in 2021 and 2022. Convert tree cover loss to proportion of total tree cover.
  2. Calculate z-score of proportion of tree cover loss
  3. Convert z-score to risk score, where...
     1. Risk 1 (lowest risk): z less than 1
     2. Risk 2: z between -1 and 0.5
     3. Risk 3: z between -0.5 and and 0.5
     4. Risk 4: z between 0.5 and 1
     5. Risk 5 (highest risk): z greater than 1
• Past Deforestation Score: The past risk score reflects the overall amount of tree cover lost at any time between 2000 and 2022.
  1. Calculate total tree cover loss from 2000 to 2022. Convert tree cover loss to proportion of total tree cover.
  2. Calculate z-score of proportion of tree cover loss.
  3. Convert z-score to risk score (same as current risk score).
• Future Deforestation Risk Score: The future risk score looks at the current risk score and the amount of remaining tree cover area (weighted equally).
  1. Calculate the current risk score z-score, as described above.
  2. Calculate the remaining proportion of tree cover. Calculate z-score of the remaining proportion of tree cover.
  3. Average the current z-score and remaining forest z-score.
  4. Convert the averaged z-score to risk score (same as above).

• Road Network: We use the Daylight Map Distribution of OpenStreetMap data as a global dataset of road networks.
• Forest Change: We access the Global Forest Change dataset via Google Earth Engine.
• Mill Locations: We use the Universal Mill List by WRI / Global Forest Watch.
• Global Coastline: To clean up geographic data, we use global coastline data provided by GADM

• In our methodology, tree cover loss and mill catchment areas are exclusive to one mill. This means that only one mill is assigned forest loss in a given area. This helps to clarify mill impact. However, multiple neighboring mills may have impact on a given forest area. Future iterations of PalmWatch may include both inclusive and exclusive catchment areas.
• Mill catchment areas are estimates of impacted areas, but we are limited in our ability to ground truth the areas of impact. Especially at a global scale, we are aware of no data to compare against our findings. We invite any feedback at palmwatch@inclusivedevelopment.net.
• Universal Mill List and brand disclosure data area dynamic. New data may become available over time and reveal different insights about consumer brand impact.
• PDF data scraping does not always fully match data, depending on the PDF formatting provided. Due to challenges in translating PDFs at scale to data tables, we cannot always match stated mill providers to known UML Listings.

Please reach out to palmwatch@inclusivedevelopment.net for general questions, comments, and media inquiries.

For more information on this project's open source code, data, and methods, please see the GitHub repository.

Find more information about the PalmWatch project partners, please see their websites at Inclusive Development International and University of Chicago Data Science Institute. This project collaboration is supported by the 11th Hour Project, a program of The Schmidt Family Foundation.

• In our methodology, tree cover loss and mill catchment areas are exclusive to one mill. This means that only one mill is assigned forest loss in a given area. This helps to clarify mill impact. However, multiple neighboring mills may have impact on a given forest area, and multiple agricultural industries may be operating in that same area. Future iterations of PalmWatch may include both inclusive and exclusive catchment areas.
• Mill catchment areas are estimates of impacted areas, but we are limited in our ability to ground truth the areas of impact. Especially at a global scale, we are aware of no data to compare against our findings. We invite any feedback at palmwatch@inclusivedevelopment.net.
• Universal Mill List and brand disclosure data are dynamic. New data may become available over time and reveal different insights about consumer brand impact. The Universal Mill List is a snapshot in time, and may not represent all the mills that are operating, or have ceased operating, in an area.
• PDF data scraping does not always fully match data, depending on the PDF formatting provided. Due to challenges in translating PDFs at scale to data tables, we cannot always match stated mill providers to known UML Listings. Brand disclosures are dynamic, and some brands may have newer disclosures that we have not yet integrated into our data processing.
• Deforestation scores are relative, not absolute. This means that additional data (new brands, new years of data) may change a brand's score.
• The Hansen tree cover loss (deforestation) data set contains tree cover loss since the year 2000. We do not account for tree cover gain, and re-loss in this methodology. Reforestation initiatives may be a mitigating effect in deforestation.

We're excited for future extensions of PalmWatch. Our website allows for community contributions to mill, brand, country, and other pages to help log events, news, and legal information.

Additional directions for future research include:

• Illuminating the Supply Chain. There are many shadow groups and middlemen traders that are missing from our data source, the Universal Mill List. We could collaborate with our partners to curate a list of additional actors in the supply chain and then add this to the website.
• Human Rights. We could elevate human rights data shared by grassroots organizations from supplementary tags to a larger website feature that permitted user search and showcased analyses of the information. Data sources could include lawsuits, complaints to non-judicial mechanisms, evidence of human rights violations gathered by local NGOs, etc.
• Mill Legality. We could investigate new data sources that would allow us to determine whether active mills had environmental permits or were operating illegally.
• Improved Methodology. We could replicate an academic study (Gaveau et. al., 2021) of deforestation in Indonesia as a best practice and to compare our findings. We could also expand the study’s methodology to new territories (e.g., South America, Africa).

The full data of mills impact by year is available at the link below:

Download Mills Data (CSV)

Universal Mill List re-publication:

• UML ID: Universal Mill List ID
• Group Name: Associated group of mills
• Parent Company: Parent company that owns the mill
• Mill Name: Name of the mill
• RSPO Status: Roundtable on Sustainable Palm Oil status, certified, not certified, or unknown
• RSPO Type: Type of RSPO certification
• Date RSPO Certification Status: Date of RSPO certification
• Latitude: Coordinate in WGS84 projection
• Longitude: Coordinate in WGS84 projection
• GPS coordinates: Coordinate string in WGS84 projection
• ISO: Three letter country code (eg. IDN)
• Country: Country name.
• Province: Province or county name.
• District: District or city name.
• Confidence level: UML verification confidence, 1 (Fully Verified), 2 (High Confidence) or 3 (Low Confidence)
• Alternative name: Other/alternative mill name, when applicable
  

Mill catchment area and tree cover loss:

• km_area: Area of the geography in square kilometers
• ha_forest_area_00 / km_forest_area_00: Total tree cover area as of 2000 in hectares or square kilometers
• treeloss_ha_2000-2022 / treeloss_km_2000-2022: Tree loss area per year (2000, 2001, etc.) in hectares or square kilometers. Each column is one year of loss
• sum_of_treeloss_km: Total tree cover loss as of 2022 in square kilometers
• treeloss_sum_proportion_of_forest: The percentage of total tree cover lost from 2000 to 2022
• remaining_proportion_of_forest: The percentage of remaining tree cover from 2000 as of 2022
• km_0, ha_0: Deprecated columns, no longer used
  

Risk score:

• risk_score_current: The current risk score based on tree cover loss in 2021 and 2022.
• risk_score_past: The past risk score based on total tree cover loss between 2000 and 2022.
• risk_score_future: A combination of the current risk score and a risk based on the remaining tree cover proportion.

Consumer brand data:

• years: A list of years the mill appeared in disclosures for that brand.

• Cyrus, C., Chandler, E., Lim, S., Jaisha, S., 2021. Palm Watch: Deforestation Tracker v0.3.
• Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L., Justice, C.O., Townshend, J.R.G., 2013. High-Resolution Global Maps of 21st-Century Forest Cover Change. Science 342, 850–853. https://doi.org/10.1126/science.1244693
• OpenStreetMap contributors, 2017. Planet dump retrieved from https://planet.osm.org.
• Universal Mill List [WWW Document], 2023. URL https://data.globalforestwatch.org/documents/gfw::universal-mill-list/about (accessed 7.31.23).