Simplifying soy deforestation q4 2025

Florian Gollnow 2026-01-30

Soy deforestation simplification

This reports analyses the potential for simplifying soy deforestation calculations avoiding:

modifying original data by using clump and eliminate removing soy plots up to 20ha
‘normalizing’ soy deforestation estimates based on IBGE production data (increase of soy deforestation beyond observed ha is questionable)

Following the Proposal shared in the Trase Data, Impact and R&D team :

keep data as is
do not modify observed deforestation, but embed the data as observed

Analysis

Compare soy area between IBGE and Glad observations
Compare new soy deforestation per ton between Normalized 20Ha clump and eliminate and Original Glad data (no-normalization, no clump))
Compare previous and new data, with original vs normalized soy deforestation data
Compare total soy deforestation between old and new data

Compare soy area between IBGE and Glad observations

Comparing GLAD-MB original, vs GLAD 20ha filtered to IBGE planted area

# A tibble: 11 × 3
    year intercept_orig intercept_20ha
   <int>          <dbl>          <dbl>
 1  2014         -126.          -332. 
 2  2015         -123.          -324. 
 3  2016         -154.          -365. 
 4  2017           40.1         -186  
 5  2018          -47.0         -244. 
 6  2019         -110.          -341. 
 7  2020          -74.5         -294. 
 8  2021          318.            53.8
 9  2022           74.7         -306. 
10  2023          423.            77.7
11  2024          225.           -58.3

Glad compares more closely to IBGE planted areas than the 20ha clump and eliminate version at National and Municpality level for the years 2014-2024. At Municipality level (see model intercepts), we find the 20ha clump and eliminate version to underestimate planted soy for all years except 2021 and 2023. Glad original (no clump and eliminate), instead often underestimates soy area until 2020 and later overestimates soy area at municiaplty level (see intercept at annual resolution).

However structural area differences within municipalities are small and all models have a high fit r^2 and are significant.

Overall, the non-clump and eliminate Glad data (SOY_GLAD_HA_orig) maps closer to the overall soy area as reported by IBGE, compared to the clump and eliminate version (SOY_GLAD_HA), suggesting to not eliminate soy areas from the mapping.

Note: This data already fills northern hemisphere regions with Mapbiomas soy areas.

Compare new soy deforestation per ton between Normalized 20Ha clump and eliminate and Original Glad data (no-normalization, no clump)

Comparing, mean, min, max, std. for those areas where IBGE reports soy harvested

[1] 0.01008469

[1] 11.04496

[1] 0.08821134

[1] 0.008054131

[1] 2.641396

[1] 0.03359501

[1] 0.009443776

[1] 3.812596

[1] 0.05395469

[1] 0.008159665

[1] 2.641396

[1] 0.03342604

We do find more outliers in soy-deforestation per ton within individual municipalities, however, these are in municipalities with little production (\<100 tons).

For visualization, production municipalities below 100tons could be marked as uncertain - or we remove per ton estimates overall.

Compare previous and new data, with original vs normalized soy deforestation data

Quantify the differences (absolute and percentage) between the new q4_2025 data and the previous q4_2023 data for overlapping years at state and country level.
A positive value indicates that the new q4_2025 data reports higher deforestation for that year than the q4_2023 data did.

Increases in absolute numbers are highest in (https://www.ibge.gov.br/explica/codigos-dos-municipios.php) Mato Grosso BR-17, Maranhao BR-29,Tocatis BR-21, Piaui BR-22, Bahia BR-31 (Matopiba)

Conclusion & Recommendations

Filtering and normalization of soy deforestation data has first reduced and second inflated trase reported soy deforestation in previous releases. While there may be argument for filtering and normalization, those arguments should not cause a distortion of the observed soy deforestation from the observed patterns based on the data - and overall soy-deforestation should be the same as the observed based on the data used. Recomendtation: do not eliminate and normalize soy deforestation, but use the data as provided.