- Activity
Tropical forest degradation
Track tree cover loss and forest degradation in tropical forests, with attribution to specific drivers.
Lead scientists
Explore the data in CTrees’ REDD+AI platform, and contact our team for access to raster data.

Detect tropical forest degradation, attributed to specific drivers
CTrees’ REDD+AI platform provides a detailed, high-resolution view of tree cover loss and forest degradation in tropical forests. First published in 2023, REDD+AI is the first and only dataset to attribute forest degradation to specific human activities: selective logging, forest fires, and road construction.
Forest degradation in tropical rainforests mainly consists of the loss of tree cover and associated carbon storage, among other ecosystem services, which does not result in complete clearing of the forest. While forest degradation can be a broad concept, we focus on the most pervasive human-induced disturbances in tropical forests that are traceable via satellite imagery.
Data in CTrees’ REDD+AI platform is distributed under the same license as the NICFI data set. This license allows non-commercial use of the data. For more information, contact CTrees (opens in a new tab).
Data at a glance
- Spatial resolution
4.77 meters for all layers
- Geographic coverage
Global tropical moist forests
- Temporal frequency and domain
- Degradation layers: bi-annual, 2017-2024
- Tree cover loss: monthly, September 2020-December 2024
- Data type & scale, offset
Type: 8-bit unsigned integer
Scale: 1
Offset: 0
- File format
20x20-kilometer tiled cloud-optimized GeoTIFF (COG)
Note: Current output is based on the NICFI Planet collection, which was discontinued after 2024. Future updates can be made using the same model and user-provided sources of hi-res satellite images.
Methodology
Input data
REDD+AI is based on Planet satellite images over the tropics made available by Norway’s International Climate and Forest Initiative (NICFI) to help save the world's tropical forests while improving the livelihoods of forest communities (Planet 2017). The Planet NICFI images are multispectral satellite images containing red, green, blue, and near-infrared bands at 4.77 x 4.77-meter spatial resolution for the Normalized Analytic Basemaps. The temporal resolution of the NICFI images is one month (Table 1), and the images are a mosaic composite of the best daily acquisitions during the month. Consequently, Planet NICFI images are mostly cloud free, thus providing a high-quality multispectral dataset to monitor land use and land cover (LULC) changes in tropical regions. Each Planet NICFI tile corresponds to an area of 20 x 20 km. More than 61,000 Planet NICFI tiles overlap the area of interest, totalling more than 24.4 million km² of analyzed area. Imagery from January 2025 was used only for confirmation of detections.
Modeling approach
The first step to the creation of the map to monitor changes is to segment the object of interest (e.g., tree cover, logging, fire, and roads), and to mask additional objects (e.g., clouds water) in the Planet NICFI images. A classical U-Net model (Ronnenberger et al, 2015) is used, which is a robust deep learning model for semantic segmentation tasks. The model was chosen because of its proven accuracy for the task (Wagner et al., 2019; Wagner et al., 2023; Dalagnol et al., 2023) and because it is fast enough for training and inference, which enables us to scale the model to the pantropics.
We developed a data processing pipeline using Amazon Web Services (AWS) cloud tools to efficiently scale our data products. The Planet NICFI dataset for our area of interest consisted of over 100 TB of compressed 8-bit data. To generate the final maps, we employed and integrated eight different AI models, including tree cover, degradation, cloud, and water cover mapping.
Validation approach and results
Validation of the datasets was performed over the Brazilian state of Mato Grosso and showed greater than 70% accuracy (Dalagnol et al., 2023). The result exceeds the performance of other machine learning-based data products and at the same time provides attribution of specific drivers of forest degradation, which is a unique feature of REDD+AI.
Pantropic-level validation is yet to be performed and is expected in Q2 2026. It will be done following good practices of map validation using a design-based stratified random sampling approach (Olofsson et al., 2014; Olofsson, 2021).
Detecting threats to forests, driving action to protect them
Applications of REDD+AI include identifying the location, date, and area of forest degradation, with attribution to specific drivers.
Forest degradation location and date
REDD+AI can be used to identify the location and date of forest degradation caused by selective logging, forest fire, and road construction occurring in the tropics. An example is shown over the state of Mato Grosso, Brazil, showing large areas affected by selective logging. When intersected with land tenure and forest management permits, REDD+AI data can be an input to assess whether activities are planned (legal) or unplanned (illegal).
Quantification of area and impacts
The data allow quantification of area of impact in tropical forests, to identify the extent and type of degradation. The area estimates can be used as input to calculate impacts such as carbon emissions resulting from forest fires in Pará, Brazil, which degraded large areas of forest in 2023 and 2024.
Attribution: Disentangling the effects of deforestation and forest degradation
The data can be used to help identify forest areas that experience land cover change, but do not necessarily constitute deforestation. Specifically, an area or pixel can experience a temporary loss of forest cover (forest cover change) due to logging activities or forest fire, and then recover over time, but this does not result in deforestation and has much lower carbon emissions than deforestation. REDD+AI is currently the only dataset in the world to attribute forest degradation to selective logging, forest fire and road construction.
Team
Access forest degradation data
Explore data (opens in a new tab) for any tropical forest jurisdiction in CTrees’ REDD+AI app.
Contact us (opens in a new tab) for data access and partnerships.
Page last updated Nov 16, 2026
All datasets
Stay Connected
Subscribe to our monthly newsletter and be the first to know about new data releases, research, and events.