Machine Learning & Remote Sensing for Healthy Forest

Machine learning and AI have developed a lot during the last 10 years. So far these technologies have not been used in private sector forestry programs and thus have not been integrated in the decision making process by forest companies in the program area. During the next 10 years national cutting volumes will be decreasing both in Estonia and Latvia. This means revenues in forestry are decreasing and only more efficient operating companies will survive. Forest owners need to implement modern technologies and machine learning analyses in their decision making and planning processes in order to make decision making more automated, objective and cost-efficient. So far forest software companies (IT-companies) in Estonia and Latvia have been lacking sufficient knowledge and investments to implement those new technologies. This project will help to overcome those challenges and make first practical solution for private forest management. In many national and international strategies it is stated that CO2 emissions have to decrease and CO2 binding can be increased by young forests. There are already active CO2 markets and tools for CO2 trading have also reached the forest sector. Forest owners will be selling CO2 credit based on their forests' ability to bind carbon more and more in the near future. But CO2 credits can be traded in young forests that are maintained carefully and timely. This project will develop a service that helps forest owners to detect areas where 1) cleaning and thinning of young forest or 2) commercial thinning is necessary based on current remote sensing data and machine learning (ML). Overall objective of the project is development of a fully automated ML4HealthyForest service for at least 300 private forest owners (with approximately 100,000 ha of forest land) in Estonia and Latvia, which provides objective automated forest management recommendations based on remote sensing data, modeling, ML and automated forest growth algorithms. Users will be private individuals, private companies and funds managing their forest. Currently such service is not available in Estonia and Latvia, but it would provide valuable data and is highly requested by forest owners in both countries. The need for such a service was identified through contact with existing forest companies and industry support organizations (such as Erametsaliit and The Estonian Forest and Wood Industries Association), which are current clients of Deskis and BSS, as well as through numerous interviews conducted by BSS as part of previous research and development projects. ML4HealthyForest service will pilot new innovative tools such as: - tools for planning and predicting future forest works based on existing tabular and field data; - automated predictions for future forest works based on modeling forest growth; - automated identification of the necessity for young stand and commercial thinning using remote sensing data and machine learning. New automated forest management tools will be integrated into existing forest management software services (ForestMan.eu in Estonia (owned by Deskis), ForestRadar in Latvia (owned by Baltic Satellite Service) and other GIS systems of the private forest sector) already in use in Estonia and Latvia. Cross-border cooperation is needed to achieve the project's objectives: 1. Collect sufficient amounts of ground truth data for developing machine-learning and modeling algorithms. Combining Estonian and Latvian forest data will result in sufficient data needed for both model development and model validation. 2. Both countries need the same algorithm. Forest owners have their own forest portfolios on both sides of the Estonian-Latvian border, and there is an increasing tendency to use the same software for managing cross-border forest portfolios. In this case, it is important to use the same models for both Estonian and Latvian forests. Especially since the tree species and management practices are quite similar in both countries. 3. Need for improved quality of Forest Registry data in both countries. Currently, national forest inventory databases are outdated, contain errors and are not reliable for objective decision-making. The proposed fully automated processing of the new LIDAR data will be very valuable, as it will provide completely new and accurate forest data that fully covers Latvia and Estonia. 4. Cross-border knowledge transfer for innovation between the partners. The transfer of cross-border knowledge will increase the competence of both partners, resulting in productive investments, innovative automation products, sustainable economic growth, higher competitiveness of both companies and the creation of new jobs. 5. Sharing of the client database will increase customer databases and competitiveness of both partners in the project area.

This project will develop a service that helps forest owners to detect areas where 1) cleaning and thinning of young forest or 2) commercial thinning is necessary based on current remote sensing data and machine learning (ML). Overall objective of the project is development of a fully automated ML4HealthyForest service for at least 300 private forest owners (with approximately 100,000 ha of forest land) in Estonia and Latvia, which provides objective automated forest management recommendations based on remote sensing data, modeling, ML and automated forest growth algorithms. Users will be private individuals, private companies and funds managing their forest. The following progress to create the proposed solution was made in cooperation of BSS and Eeway: 1) review of scientific literature; 2) information exchange about data necessary for the project; 3) questionnaire conducted with forest owners from Latvia and Estonia and system requirements collected; 4) Architecture design was created; 5) Technical development of REST API and user interface design has been started; 6) ML/data science and remote sensing knowledge exchange and training conducted including practical examples of ML; 7) Kick-off meeting took place in Gungas (LV) on 16.09.2024 with participation of 3 persons from BSS and 4 persons from Eeway. 8) Team-building event took place in Tartu (EE) on 2.12.2024 with participation of 2 persons from BSS and 4 persons from Eeway; 8) Project description presented on web pages of both partners. The first prototype of developed forest growth models and user interface to present it has been developed but not yet finished and tested: 1) Both partners have done a common work in discussing necessary data for the project, ground truth data, researching availability, structure, quality, containing information etc. Eeway provided comprehensive forest-related information which was discussed and analysed during the common virtual meetings and finished ground truth data collection. 2) Eeway is in the progress of developing models that validate the ground truth data and predict forest works. 3) The common list of data with descriptions was prepared for LV and EE to be used for the project needs by both partners. BSS provided validation procedure for ensuring compatibility between ground truth forest inventory information and CHM which will be the main input source for training the machine learning methods. 4) BSS has started training of ML models for single tree detection, RandomForest based and alternative workflow development with an objective being able to estimate the need for forest stand thinning. 5) BSS proceeded with development of backend REST API 6) BSS proceeded with development of JavaScript widgets in the Svelte framework for presenting forest data in a web application. 7) Development web UI for forest works and predictions is in progress. 8) Marketing activities to introduce project to forest industry, international EU FIERCE project managers, Estonian university of Life Sciences and Ministry of Education and Science (LV). Significant progress has been made by both partners in developing tools, workflows, and models to support data management, preprocessing, and LiDAR-based forest analysis. The key achievements of BSS include: - Development of Python-based tools for data management and preprocessing. - Automated .laz file download tool from the Estonian Geoportal (https://geoportaal.maaamet.ee/). - Batch conversion tool from .laz to .las format using LASTools. - Automated processing tool for batch application of FUSION utilities to calculate LiDAR point cloud features. - Conversion tool for transforming FUSION-generated CSV files into multiband GeoTIFFs. - Python workflow integration for Workflow 1 and Workflow 2 (detailed in the Technical Project Report, chapter “Methodology Description”). - Model training completed for both Workflow 1 and Workflow 2. - Validation and accuracy assessment performed on test datasets, including case studies to analyse error sources and model transferability between geographical regions. In addition, BSS has obtained comprehensive forest thinning ground truth data from the State Forest Register of Latvia, which was essential for accurate machine learning model training, development, and testing. Eeway has successfully advanced the development and integration of the ForestMan application and underlying models. Achievements include: - Completion of models validating ground truth data and finalisation of the forest growth model. - Completion of models predicting forest management operations. - Development and finalisation of the web user interface within the ForestMan system. - Initiation of testing and validation activities for the developed modules. - Start of integration of the models into the existing ForestMan system. Project partners have initiated API testing with the currently developed software modules. During the third reporting period, project activities were primarily focused on the development and integration of the new machine learning–based forest monitoring service, growth model, and user interface. As the technical implementation was still in progress, it was not yet possible to onboard private forest owners for active testing. However, forest owners were informed about the upcoming solution and its planned functionalities through presentations and communication activities. Once the service is fully implemented within the ForestRadar and ForestMan systems, forest owners will be invited to test the system and provide feedback, ensuring practical validation and user-driven improvement of the developed solution. The project has been presented at multiple events, with BSS also registered as a speaker at upcoming events in September 2025 to showcase the expected project outcomes. 1. ESA Conference , New Capabilities and countries in European Space Conference, Frascatti, Italy, May 14-15, 2025 2. Space Resources conference, Krakow, Poland, May 21-23, 2025 3. Canada–Baltics Horizon Europe Networking Event, on-line, May 29, 2025 4. Indian Space Agency meeting, on-line, June 4 5. Meeting with all representatives of Latvian Chamber of Commerce and Industries, Riga, Latvia, June 20, 2025 6. Cost Action PANGEOS Conference on Remote sensing applications towards European Green Deal strategic goals, Jurmala, Latvia, September 11-12, 2025 7. Big Data Foundations for Earth Observation 2025, Riga, Latvia, September 22-26, 2025 8. Big Data from Space Conference, Riga, Latvia , September 29 – October 3, 2025 9. SATSDISFACTION conference Use of Satellite Data for the achievement of the Sustainable Development Goals 2030, Riga, Latvia, September 10-11, 2025