🌟 IntroductionThis repository provides the data used in the research by NIBIO on active learning techniques for UAV image classification of forest damage due to wind storms. 🌲 Scope of the DataThis dataset is intended for:🔍 Development and benchmarking of image classification models for UAV RGB imagery according to the level of damage (i.e. windblown trees). ⚖️ Licensing📄 Please refer to the specific licenses below for details on how the data can be used.🔑 Key Licensing Principles:✅ You may access, use, and share the dataset and models freely.🔄 Any derivative works (e.g., trained models, code for training, or prediction tools) must also be made publicly available under the same licensing terms.🌍 These licenses promote collaboration and transparency, ensuring that research using this dataset benefits the broader scientific and open-source community 🙌

🌟 IntroductionThis repository provides the data used in the research by NIBIO on active learning techniques for UAV image classification of forest damage due to wind storms. 🌲 Scope of the DataThis dataset is intended for:🔍 Development and benchmarking of image classification models for UAV RGB imagery according to the level of damage (i.e. windblown trees). ⚖️ Licensing📄 Please refer to the specific licenses below for details on how the data can be used.🔑 Key Licensing Principles:✅ You may access, use, and share the dataset and models freely.🔄 Any derivative works (e.g., trained models, code for training, or prediction tools) must also be made publicly available under the same licensing terms.🌍 These licenses promote collaboration and transparency, ensuring that research using this dataset benefits the broader scientific and open-source community 🙌

🌟 IntroductionThis repository provides the data used in the research by Puliti and Astrup (2022) Automatic detection of snow breakage at single tree level using YOLOv5 applied to UAV imagery. International Journal of Applied Earth Observation and Geoinformation, 112, p.102946. 🌲 Scope of the DataThis dataset is intended for:🔍 Development and benchmarking of object detection models for individual trees and classification of trees based on their health.Data is provided in the YOLO format with bounding box labels 📦🌲 🖥️ Existing Code and ModelThe code for model inference, as described in the paper by Puliti and Astrup (2022), is available in the following GitHub repository:🔗 GitHub Repository for Model InferenceThis repository includes:Inference Scripts: Scripts to apply the trained YOLOv5 model for detecting snow breakage at the single-tree level. 🌲Pre-trained Models: Downloadable weights for reproducing results from the publication.Example Workflows: Step-by-step guidance for running the model on your own UAV imagery. 🚁Make sure to follow the repository’s documentation for setup instructions, dependencies, and usage examples. 💻 📜 CitationIf you use this dataset, please give credit by citing the original paper:@article{PULITI2022102946,title = {Automatic detection of snow breakage at single tree level using YOLOv5 applied to UAV imagery},journal = {International Journal of Applied Earth Observation and Geoinformation},volume = {112},pages = {102946},year = {2022},issn = {1569-8432},doi = {https://doi.org/10.1016/j.jag.2022.102946},url = {https://www.sciencedirect.com/science/article/pii/S1569843222001431},author = {Stefano Puliti and Rasmus Astrup},keywords = {Forest damage, Convolutional neural network, Deep-learning, Drones, Object detection}} ⚖️ Licensing📄 Please refer to the specific licenses below for details on how the data can be used.🔑 Key Licensing Principles:✅ You may access, use, and share the dataset and models freely.🔄 Any derivative works (e.g., trained models, code for training, or prediction tools) must also be made publicly available under the same licensing terms.🌍 These licenses promote collaboration and transparency, ensuring that research using this dataset benefits the broader scientific and open-source community 🙌

🌟 IntroductionThis repository provides the data used in the research by Puliti and Astrup (2022) Automatic detection of snow breakage at single tree level using YOLOv5 applied to UAV imagery. International Journal of Applied Earth Observation and Geoinformation, 112, p.102946. 🌲 Scope of the DataThis dataset is intended for:🔍 Development and benchmarking of object detection models for individual trees and classification of trees based on their health.Data is provided in the YOLO format with bounding box labels 📦🌲 🖥️ Existing Code and ModelThe code for model inference, as described in the paper by Puliti and Astrup (2022), is available in the following GitHub repository:🔗 GitHub Repository for Model InferenceThis repository includes:Inference Scripts: Scripts to apply the trained YOLOv5 model for detecting snow breakage at the single-tree level. 🌲Pre-trained Models: Downloadable weights for reproducing results from the publication.Example Workflows: Step-by-step guidance for running the model on your own UAV imagery. 🚁Make sure to follow the repository’s documentation for setup instructions, dependencies, and usage examples. 💻 📜 CitationIf you use this dataset, please give credit by citing the original paper:@article{PULITI2022102946,title = {Automatic detection of snow breakage at single tree level using YOLOv5 applied to UAV imagery},journal = {International Journal of Applied Earth Observation and Geoinformation},volume = {112},pages = {102946},year = {2022},issn = {1569-8432},doi = {https://doi.org/10.1016/j.jag.2022.102946},url = {https://www.sciencedirect.com/science/article/pii/S1569843222001431},author = {Stefano Puliti and Rasmus Astrup},keywords = {Forest damage, Convolutional neural network, Deep-learning, Drones, Object detection}} ⚖️ Licensing📄 Please refer to the specific licenses below for details on how the data can be used.🔑 Key Licensing Principles:✅ You may access, use, and share the dataset and models freely.🔄 Any derivative works (e.g., trained models, code for training, or prediction tools) must also be made publicly available under the same licensing terms.🌍 These licenses promote collaboration and transparency, ensuring that research using this dataset benefits the broader scientific and open-source community 🙌

DescriptionData for benchmarking tree species classification from proximally-sensed laser scanning data.Data split and usageThe data is split into:Development data (dev): these includes 90% of the trees in the dataset and consists of individual tree point clouds (.laz) named according to the treeID column available in the tree_metadata_dev.csv file, from which tree_species labels are available. These data are meant to be used for model development and can thus be further split into training and validation datasets.Test data (test): these are 10% of the trees (balanced sample) and include individual tree point clouds (.laz) but, for benchmarking purposes, the species labels are witheld for benchmarking purposes. Thus to make use of the test data the users should predict species on the test trees, and output a table (.csv file) with a row per predicted tree and two columns (treeID and predicted_species). This table can then be used to create a new submission in the FOR-species20K Codabench benchmarking platform and obtain the evaluation metrics corresponding to the test data.CiteAny scientific publication using the data should cite the following paper:Puliti, S., Lines, E., Müllerová, J., Frey, J., Schindler, Z., Straker, A., Allen, M.J., Winiwarter, L., Rehush, N., Hristova, H., Murray, B., Calders, K., Terryn, L., Coops, N., Höfle, B., Krůček, M., Krokm, G., Král, K., Luck, L., Levick, S.R., Missarov, A., Mokroš, M., Owen, H., Stereńczak, K., Pitkänen, T.P., Puletti, N., Saarinen, N., Hopkinson, C., Torresan, C., Tomelleri, E., Weiser, H., Junttila, S., and Astrup, R. (2025) Benchmarking tree species classification from proximally-sensed laser scanning data: introducing the FOR-species20K dataset. Methods in Ecology and Evolution, 00,1–18. Available here

DescriptionData for benchmarking tree species classification from proximally-sensed laser scanning data.Data split and usageThe data is split into:Development data (dev): these includes 90% of the trees in the dataset and consists of individual tree point clouds (.laz) named according to the treeID column available in the tree_metadata_dev.csv file, from which tree_species labels are available. These data are meant to be used for model development and can thus be further split into training and validation datasets.Test data (test): these are 10% of the trees (balanced sample) and include individual tree point clouds (.laz) but, for benchmarking purposes, the species labels are witheld for benchmarking purposes. Thus to make use of the test data the users should predict species on the test trees, and output a table (.csv file) with a row per predicted tree and two columns (treeID and predicted_species). This table can then be used to create a new submission in the FOR-species20K Codabench benchmarking platform and obtain the evaluation metrics corresponding to the test data.CiteAny scientific publication using the data should cite the following paper:Puliti, S., Lines, E., Müllerová, J., Frey, J., Schindler, Z., Straker, A., Allen, M.J., Winiwarter, L., Rehush, N., Hristova, H., Murray, B., Calders, K., Terryn, L., Coops, N., Höfle, B., Krůček, M., Krokm, G., Král, K., Luck, L., Levick, S.R., Missarov, A., Mokroš, M., Owen, H., Stereńczak, K., Pitkänen, T.P., Puletti, N., Saarinen, N., Hopkinson, C., Torresan, C., Tomelleri, E., Weiser, H., Junttila, S., and Astrup, R. (2025) Benchmarking tree species classification from proximally-sensed laser scanning data: introducing the FOR-species20K dataset. Methods in Ecology and Evolution, 00,1–18. Available here

We developed Pan-European maps of timber volume (V), above-ground biomass (AGB), and deciduous-coniferous proportion (DCP) with a pixel size of 10 x 10 m2 for the reference year 2020 using a combination of a Sentinel 2 mosaic, Copernicus layers, and National Forest Inventory (NFI) data.For mapping, we used the k-Nearest Neighbor (kNN, k=7) approach with a harmonized database of species-specific V and AGB from 14 NFIs across Europe. This database encompasses approximately 151,000 sample plots, which were intersected with the above-mentioned Earth observation data. The maps cover 40 European countries, forming a continuous coverage of the western part of the European continent.A sample of 1/3 of NFI plots was left out for validation, whereas 2/3 of the plots were used for mapping. Maps were created independently for 13 multi-country processing areas. Root-mean-squared-errors (RMSEs) for AGB ranged from 53 % in the Nordic processing area to 73 % the South-Eastern area.The created maps are the first of their kind as they are utilizing a huge amount of harmonized NFI observations and consistent remote sensing data for high-resolution forest attribute mapping. While the published maps can be useful for visualization and other purposes, they are primarily meant as auxiliary information in model-assisted estimation where model-related biases can be mitigated, and field-based estimates improved. Therefore, additional calibration procedures were not applied, and especially high V and AGB values tend to be underestimated. Summarizing map values (pixel counting) over large regions such as countries or whole Europe will consequently result in biased estimates that need to be interpreted with care.The author list is sorted by last name except for the first and last authors who also serve as corresponding authors.Corresponding authors: [email protected], [email protected]

We developed Pan-European maps of timber volume (V), above-ground biomass (AGB), and deciduous-coniferous proportion (DCP) with a pixel size of 10 x 10 m2 for the reference year 2020 using a combination of a Sentinel 2 mosaic, Copernicus layers, and National Forest Inventory (NFI) data.For mapping, we used the k-Nearest Neighbor (kNN, k=7) approach with a harmonized database of species-specific V and AGB from 14 NFIs across Europe. This database encompasses approximately 151,000 sample plots, which were intersected with the above-mentioned Earth observation data. The maps cover 40 European countries, forming a continuous coverage of the western part of the European continent.A sample of 1/3 of NFI plots was left out for validation, whereas 2/3 of the plots were used for mapping. Maps were created independently for 13 multi-country processing areas. Root-mean-squared-errors (RMSEs) for AGB ranged from 53 % in the Nordic processing area to 73 % the South-Eastern area.The created maps are the first of their kind as they are utilizing a huge amount of harmonized NFI observations and consistent remote sensing data for high-resolution forest attribute mapping. While the published maps can be useful for visualization and other purposes, they are primarily meant as auxiliary information in model-assisted estimation where model-related biases can be mitigated, and field-based estimates improved. Therefore, additional calibration procedures were not applied, and especially high V and AGB values tend to be underestimated. Summarizing map values (pixel counting) over large regions such as countries or whole Europe will consequently result in biased estimates that need to be interpreted with care.The author list is sorted by last name except for the first and last authors who also serve as corresponding authors.Corresponding authors: [email protected], [email protected]

General descriptionThis dataset consists of a ML-ready labelled mobile laser scanning (MLS) point cloud dataset including 16 manually labelled forest plots (approx. 250 m2) for forest panoptic segmentation and thus including both semantic and instance labels. The data was collected using a Geoslam Horizon RT and processed using Geoslam Hub.LabelsThe data were then labelled into the following semantic classes (label):1= ground2= vegetation: these include both branches, leaves, and low vegetation3= lying deadwood4= stemsIn addition for each tree, a unique tree identifier (treeID) was also assigned to each point.Data splitEach plot was split into train (50%), validation (25%), and test (25%) sets by dividing the circular plot into four slices, out of which the first two were used for training, the third for validation, and the fourth for test. Thus the users might play around with merging the train and validation dataset as they prefer. These two sets can be used during model training, hyperparameter tuning, and model selection. However, the test set should be kept as an independent set to be used for benchmarking against the values reported in the two studies indicated below. CitationTo cite this datasets and for a more detailed description use:Wielgosz, M., Puliti, S., Xiang, B., Schindler, K. and Astrup, R., 2024. SegmentAnyTree: A sensor and platform agnostic deep learning model for tree segmentation using laser scanning data. Remote Sensing of Environment; Other studies using these dataWielgosz, M., Puliti, S., Wilkes, P. and Astrup, R., 2023. Point2Tree (P2T)—Framework for parameter tuning of semantic and instance segmentation used with mobile laser scanning data in coniferous forest. Remote Sensing, 15(15), p.3737; available hereFundingThis work is part of the Center for Research-based Innovation SmartForest: Bringing Industry 4.0 tothe Norwegian forest sector (NFR SFI project no. 309671, smartforest.no).⚖️ Licensing📄 Please refer to the specific licenses below for details on how the data can be used.🔑 Key Licensing Principles:✅ You may access, use, and share the dataset and models freely.🔄 Any derivative works (e.g., trained models, code for training, or prediction tools) must also be made publicly available under the same licensing terms.🌍 These licenses promote collaboration and transparency, ensuring that research using this dataset benefits the broader scientific and open-source community 🙌

General descriptionThis dataset consists of a ML-ready labelled mobile laser scanning (MLS) point cloud dataset including 16 manually labelled forest plots (approx. 250 m2) for forest panoptic segmentation and thus including both semantic and instance labels. The data was collected using a Geoslam Horizon RT and processed using Geoslam Hub.LabelsThe data were then labelled into the following semantic classes (label):1= ground2= vegetation: these include both branches, leaves, and low vegetation3= lying deadwood4= stemsIn addition for each tree, a unique tree identifier (treeID) was also assigned to each point.Data splitEach plot was split into train (50%), validation (25%), and test (25%) sets by dividing the circular plot into four slices, out of which the first two were used for training, the third for validation, and the fourth for test. Thus the users might play around with merging the train and validation dataset as they prefer. These two sets can be used during model training, hyperparameter tuning, and model selection. However, the test set should be kept as an independent set to be used for benchmarking against the values reported in the two studies indicated below. CitationTo cite this datasets and for a more detailed description use:Wielgosz, M., Puliti, S., Xiang, B., Schindler, K. and Astrup, R., 2024. SegmentAnyTree: A sensor and platform agnostic deep learning model for tree segmentation using laser scanning data. Remote Sensing of Environment; Other studies using these dataWielgosz, M., Puliti, S., Wilkes, P. and Astrup, R., 2023. Point2Tree (P2T)—Framework for parameter tuning of semantic and instance segmentation used with mobile laser scanning data in coniferous forest. Remote Sensing, 15(15), p.3737; available hereFundingThis work is part of the Center for Research-based Innovation SmartForest: Bringing Industry 4.0 tothe Norwegian forest sector (NFR SFI project no. 309671, smartforest.no).⚖️ Licensing📄 Please refer to the specific licenses below for details on how the data can be used.🔑 Key Licensing Principles:✅ You may access, use, and share the dataset and models freely.🔄 Any derivative works (e.g., trained models, code for training, or prediction tools) must also be made publicly available under the same licensing terms.🌍 These licenses promote collaboration and transparency, ensuring that research using this dataset benefits the broader scientific and open-source community 🙌