Objectives of the service
Timber operators must comply with the European Union Deforestation Regulation (EUDR), ensuring products are deforestation-free and do not cause forest degradation. However, current satellite systems frequently misidentify legal timber harvesting as illegal deforestation, creating immense manual review burdens. Furthermore, the industry lacks scientific consensus on measuring degradation, and user-supplied geospatial boundaries (polygons) often contain formatting errors.
To solve this, NGIS will deliver ForestHD, a dedicated module for its existing TraceMark Software as a Service (SaaS) platform. Powered by Earth Observation (EO) data and Machine Learning (ML), ForestHD automatically corrects faulty polygon data, accurately distinguishes legal harvests from true deforestation to eliminate false alarms, and provides scientific forest degradation metrics. The polygon cleaning tool will also be offered as a standalone Application Programming Interface (API) micro-service.
The project's overarching goal is to successfully develop, validate, and commercialize these tools. Key activities include gathering user requirements, generating robust training datasets using satellite and timber company data, building the classification models, and fully integrating the final module into TraceMark through rigorous testing with industry pilot users.
Users and their needs
The primary targeted user communities are timber operators, paper and cardboard producers, EU importers, and forestry certifying bodies like PEFC and FSC. The specific users currently involved in testing the service are based in North America (the United States and Canada).
User Needs:
- Prove commodities are deforestation-free and degradation-free for EUDR compliance.
- Automate the cleaning of massive volumes of irregular, field-collected geospatial polygon data.
- Eliminate the unfeasible costs and operational burdens associated with manual ground inspections.
- Monitor certified forestry areas to maintain standards, carbon credit validity, and brand reputation.
Project Challenges to Meet Needs: The project faces significant technical challenges to fulfil these needs. Primarily, it must successfully train Machine Learning models using Earth Observation data to accurately distinguish between legal timber harvesting and true deforestation, thereby minimizing false positive alerts. Furthermore, it must develop a robust automated tool capable of fixing widespread geometry errors in user-submitted polygons, and successfully establish validated, scientific metrics to quantify forest degradation at a landscape scale.
Service/ system concept
User Capabilities and Supplied Information
ForestHD provides users with three core capabilities for EUDR compliance:
- Polygon Data Cleaning: Automatically corrects formatting and drawing errors in uploaded raw geospatial data (such as GPS boundary tracks) to ensure valid mapping of suppliers farms.
- Harvesting vs. Deforestation Detection: Uses smart algorithms to accurately distinguish legal timber harvests from illegal deforestation events at a plot level, significantly reducing false positive alerts.
- Forest Degradation Metrics: Measures and estimates long-term forest health and degradation across the sourcing landscape.
Ultimately, users receive verified, scientifically-backed data to easily prove compliance and share reports with downstream clients and auditors.
How the System Works
The system is designed for ease of use without technical mapping skills.
- Upload: Users log into the secure TraceMark dashboard to upload their field data.
- Clean: The polygon cleaning tool automatically fixes mapping errors so boundaries are perfectly aligned.
- Analyze: Cleaned data is sent to the Google Cloud Platform. Here, machine learning algorithms analyze the boundaries against Earth Observation satellite imagery to detect legal harvesting, illegal clearing, and overall degradation.
- Report: The system generates a simple, accurate compliance report
Space Added Value
ForestHD utilizes several key space assets: Copernicus Sentinel-1 GRD and Sentinel-2 L2A satellite imagery, Digital Elevation Models (e.g., SRTM), and satellite positioning data from customers recording the navigation tracks of harvesting machinery.
Expected Added Value vs. Competitors: Existing competitor platforms primarily base their alerts on basic tree-loss detection. This creates a massive operational burden for the timber industry, as legal, sustainable harvesting is constantly misidentified as illegal deforestation, resulting in unsustainable false positive alarms.
By combining continuous optical and radar Earth Observation (EO) imagery with ground-level satellite navigation tracks, ForestHD trains advanced Machine Learning (ML) models to accurately distinguish between true deforestation and legal harvesting. Furthermore, combining these space assets allows the system to measure continuous, long-term forest degradation at a landscape scale—a complex metric currently missing from competitor platforms. Ultimately, fusing these space assets delivers precise, transparent compliance reporting that eliminates the false alarms and "black box" limitations of current existing solutions.
Current Status
The ForestHD project is currently making strong progress through its initial phases. A major focus has been on gathering precise user requirements by hosting collaborative workshops with our pilot users, which include forestry industry leaders like Domtar and international certification bodies like PEFC. These interactive workshops have been critical for aligning the system's design with the real-world operational challenges timber operators face regarding EUDR compliance.
Concurrently, technical work has officially started on the automated Polygon Cleaning Tool. The development team is actively implementing the quality control checks to identify and resolve common formatting and geometry errors in field-collected data, such as self-intersections, coordinate system discrepancies, and invalid shapes. This foundational development will ensure the system can seamlessly ingest and correct complex geospatial boundaries before we move forward with training the advanced machine learning models.
