Overview: 321: Grasslands under no or moderate human influence. Low productivity grasslands. Often situated in areas of rough, uneven ground, steep slopes; frequently including rocky areas or patches of other (semi-)natural vegetation. Natural grasslands are areas with herbaceous vegetation (maximum height is 150 cm and gramineous species are prevailing) covering at least 50 % of the surface. Besides herbaceous vegetation, areas of shrub formations, of scattered trees and of mineral outcrops also occur. Often under nature conservation. In this context the term ”natural” indicates that vegetation is developed under a minimum human interference,(not mowed, drained, irrigated, sown, fertilized or stimulated by chemicals, which might influence production of biomass). Even though the human interference cannot be completely discarded in quoted areas, it does not suppress the natural development or species composition of the meadows. Maintenance mowing and shrub clearance for prevention of woody overgrowth due to natural succession is tolerated. Sporadic extensive grazing with low livestock unit/ha is possible. Typical visible characteristics: large extent, irregular shape, usually in distant location from larger settlements. Traceability (lineage): This dataset was produced with a machine learning framework with several input datasets, specified in detail in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ) Scientific methodology: The single-class probability layers were generated with a spatiotemporal ensemble machine learning framework detailed in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ). The single-class uncertainty layers were calculated by taking the standard deviation of the three single-class probabilities predicted by the three components of the ensemble. The HCL (hard class) layers represents the class with the highest probability as predicted by the ensemble. Usability: The HCL layers have a decreasing average accuracy (weighted F1-score) at each subsequent level in the CLC hierarchy. These metrics are 0.83 at level 1 (5 classes):, 0.63 at level 2 (14 classes), and 0.49 at level 3 (43 classes). This means that the hard-class maps are more reliable when aggregating classes to a higher level in the hierarchy (e.g. 'Discontinuous Urban Fabric' and 'Continuous Urban Fabric' to 'Urban Fabric'). Some single-class probabilities may more closely represent actual patterns for some classes that were overshadowed by unequal sample point distributions. Users are encouraged to set their own thresholds when postprocessing these datasets to optimize the accuracy for their specific use case. Uncertainty quantification: Uncertainty is quantified by taking the standard deviation of the probabilities predicted by the three components of the spatiotemporal ensemble model. Data validation approaches: The LULC classification was validated through spatial 5-fold cross-validation as detailed in the accompanying publication. Completeness: The dataset has chunks of empty predictions in regions with complex coast lines (e.g. the Zeeland province in the Netherlands and the Mar da Palha bay area in Portugal). These are artifacts that will be avoided in subsequent versions of the LULC product. Consistency: The accuracy of the predictions was compared per year and per 30km*30km tile across europe to derive temporal and spatial consistency by calculating the standard deviation. The standard deviation of annual weighted F1-score was 0.135, while the standard deviation of weighted F1-score per tile was 0.150. This means the dataset is more consistent through time than through space: Predictions are notably less accurate along the Mediterrranean coast. The accompanying publication contains additional information and visualisations. Positional accuracy: The raster layers have a resolution of 30m, identical to that of the Landsat data cube used as input features for the machine learning framework that predicted it. Temporal accuracy: The dataset contains predictions and uncertainty layers for each year between 2000 and 2019. Thematic accuracy: The maps reproduce the Corine Land Cover classification system, a hierarchical legend that consists of 5 classes at the highest level, 14 classes at the second level, and 44 classes at the third level. Class 523: Oceans was omitted due to computational constraints.

122: Motorways and railways, including associated installations (stations, platforms, embankments, linear greenery narrower than 100 m). Minimum width for inclusion: 100 m. The general requirement of 100 m delineation accuracy is not sufficient in mapping 122. The tolerable shift in delineation is maximum 50 m. In delineating 122 a maximum 15-20% exaggeration of width is allowed, meaning that real width of the road including associated land should be at least 80 m to be included in CLC. In such cases the exaggerated width should be as close as possible to 100 m.

Overview: Potential Natural Vegetation (PNV): potential probability of occurrence for the Turkey oak from 2018 to 2020 Traceability (lineage): This is an original dataset produced with a machine learning framework which used a combination of point datasets and raster datasets as inputs. Point dataset is a harmonized collection of tree occurrence data, comprising observations from National Forest Inventories (EU-Forest), GBIF and LUCAS. The complete dataset is available on Zenodo. Raster datasets used as input are: monthly time series air and surface temperature and precipitation from a reprocessed version of the Copernicus ERA5 dataset; long term averages of bioclimatic variables from CHELSA; elevation, slope and other elevation-derived metrics and long term monthly averages snow probability. For a more comprehensive list refer to Bonannella et al. (2022) (in review, preprint available at: https://doi.org/10.21203/rs.3.rs-1252972/v1). Scientific methodology: Probability and uncertainty maps were the output of a spatiotemporal ensemble machine learning framework based on stacked regularization. Three base models (random forest, gradient boosted trees and generalized linear models) were first trained on the input dataset and their predictions were used to train an additional model (logistic regression) which provided the final predictions. More details on the whole workflow are available in the listed publication. Usability: Probability maps are particularly useful when compared with existing products of potential distribution of species or when combined with maps of realized distribution: gaps in potential and realized distribution can be identified and used as information for future programs of tree planting or forest restoration. Uncertainty quantification: Uncertainty is quantified by taking the standard deviation of the probabilities predicted by the three components of the spatiotemporal ensemble model. Data validation approaches: Distribution maps were validated using a spatial 5-fold cross validation following the workflow detailed in the listed publication. Completeness: The raster files perfectly cover the entire Geo-harmonizer region as defined by the landmask raster dataset available here. Consistency: Areas which are outside of the calibration area of the point dataset (Iceland, Norway) usually have high uncertainty values. This is not only a problem of extrapolation but also of poor representation in the feature space available to the model of the conditions that are present in this countries. Positional accuracy: The rasters have a spatial resolution of 30m. Temporal accuracy: The maps cover the period 2018 - 2020 Thematic accuracy: Both probability and uncertainty maps contain values from 0 to 100: in the case of probability maps, they indicate the probability of occurrence of a single individual of the target species, while uncertainty maps indicate the standard deviation of the ensemble model.

Overview: 333: Areas with sparse vegetation, covering 10-50% of surface. Includes steppes, tundra, lichen heath, badlands, karstic areas and scattered high-altitude vegetation. Scattered vegetation is composed of herbaceous and/or ligneous and semi-ligneous species, the rest of area is naturally bare ground. Traceability (lineage): This dataset was produced with a machine learning framework with several input datasets, specified in detail in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ) Scientific methodology: The single-class probability layers were generated with a spatiotemporal ensemble machine learning framework detailed in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ). The single-class uncertainty layers were calculated by taking the standard deviation of the three single-class probabilities predicted by the three components of the ensemble. The HCL (hard class) layers represents the class with the highest probability as predicted by the ensemble. Usability: The HCL layers have a decreasing average accuracy (weighted F1-score) at each subsequent level in the CLC hierarchy. These metrics are 0.83 at level 1 (5 classes):, 0.63 at level 2 (14 classes), and 0.49 at level 3 (43 classes). This means that the hard-class maps are more reliable when aggregating classes to a higher level in the hierarchy (e.g. 'Discontinuous Urban Fabric' and 'Continuous Urban Fabric' to 'Urban Fabric'). Some single-class probabilities may more closely represent actual patterns for some classes that were overshadowed by unequal sample point distributions. Users are encouraged to set their own thresholds when postprocessing these datasets to optimize the accuracy for their specific use case. Uncertainty quantification: Uncertainty is quantified by taking the standard deviation of the probabilities predicted by the three components of the spatiotemporal ensemble model. Data validation approaches: The LULC classification was validated through spatial 5-fold cross-validation as detailed in the accompanying publication. Completeness: The dataset has chunks of empty predictions in regions with complex coast lines (e.g. the Zeeland province in the Netherlands and the Mar da Palha bay area in Portugal). These are artifacts that will be avoided in subsequent versions of the LULC product. Consistency: The accuracy of the predictions was compared per year and per 30km*30km tile across europe to derive temporal and spatial consistency by calculating the standard deviation. The standard deviation of annual weighted F1-score was 0.135, while the standard deviation of weighted F1-score per tile was 0.150. This means the dataset is more consistent through time than through space: Predictions are notably less accurate along the Mediterrranean coast. The accompanying publication contains additional information and visualisations. Positional accuracy: The raster layers have a resolution of 30m, identical to that of the Landsat data cube used as input features for the machine learning framework that predicted it. Temporal accuracy: The dataset contains predictions and uncertainty layers for each year between 2000 and 2019. Thematic accuracy: The maps reproduce the Corine Land Cover classification system, a hierarchical legend that consists of 5 classes at the highest level, 14 classes at the second level, and 44 classes at the third level. Class 523: Oceans was omitted due to computational constraints.

nuts: Administrative areas according to NUTS and OSM, rasterized to provide one single hierarchical code as pixel value for each county in EU.

223: Cultivated areas planted with olive trees.

332: Scree, cliffs, rock outcrops, including areas of active erosion, rocks and reef flats situated above the high-water mark, inland salt planes.

Overview: 121: Land units that are under industrial or commercial use or serve for public service facilities. Traceability (lineage): This dataset was produced with a machine learning framework with several input datasets, specified in detail in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ) Scientific methodology: The single-class probability layers were generated with a spatiotemporal ensemble machine learning framework detailed in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ). The single-class uncertainty layers were calculated by taking the standard deviation of the three single-class probabilities predicted by the three components of the ensemble. The HCL (hard class) layers represents the class with the highest probability as predicted by the ensemble. Usability: The HCL layers have a decreasing average accuracy (weighted F1-score) at each subsequent level in the CLC hierarchy. These metrics are 0.83 at level 1 (5 classes):, 0.63 at level 2 (14 classes), and 0.49 at level 3 (43 classes). This means that the hard-class maps are more reliable when aggregating classes to a higher level in the hierarchy (e.g. 'Discontinuous Urban Fabric' and 'Continuous Urban Fabric' to 'Urban Fabric'). Some single-class probabilities may more closely represent actual patterns for some classes that were overshadowed by unequal sample point distributions. Users are encouraged to set their own thresholds when postprocessing these datasets to optimize the accuracy for their specific use case. Uncertainty quantification: Uncertainty is quantified by taking the standard deviation of the probabilities predicted by the three components of the spatiotemporal ensemble model. Data validation approaches: The LULC classification was validated through spatial 5-fold cross-validation as detailed in the accompanying publication. Completeness: The dataset has chunks of empty predictions in regions with complex coast lines (e.g. the Zeeland province in the Netherlands and the Mar da Palha bay area in Portugal). These are artifacts that will be avoided in subsequent versions of the LULC product. Consistency: The accuracy of the predictions was compared per year and per 30km*30km tile across europe to derive temporal and spatial consistency by calculating the standard deviation. The standard deviation of annual weighted F1-score was 0.135, while the standard deviation of weighted F1-score per tile was 0.150. This means the dataset is more consistent through time than through space: Predictions are notably less accurate along the Mediterrranean coast. The accompanying publication contains additional information and visualisations. Positional accuracy: The raster layers have a resolution of 30m, identical to that of the Landsat data cube used as input features for the machine learning framework that predicted it. Temporal accuracy: The dataset contains predictions and uncertainty layers for each year between 2000 and 2019. Thematic accuracy: The maps reproduce the Corine Land Cover classification system, a hierarchical legend that consists of 5 classes at the highest level, 14 classes at the second level, and 44 classes at the third level. Class 523: Oceans was omitted due to computational constraints.

osm: Farmyard rasterized from OSM landuse polygons, first to 10m spatial resolution and after downsampled to 30m by spatial average.

Overview: 243: Areas principally occupied with agriculture, interspersed with significantsemi-natural areas in a mosaic pattern. Traceability (lineage): This dataset was produced with a machine learning framework with several input datasets, specified in detail in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ) Scientific methodology: The single-class probability layers were generated with a spatiotemporal ensemble machine learning framework detailed in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ). The single-class uncertainty layers were calculated by taking the standard deviation of the three single-class probabilities predicted by the three components of the ensemble. The HCL (hard class) layers represents the class with the highest probability as predicted by the ensemble. Usability: The HCL layers have a decreasing average accuracy (weighted F1-score) at each subsequent level in the CLC hierarchy. These metrics are 0.83 at level 1 (5 classes):, 0.63 at level 2 (14 classes), and 0.49 at level 3 (43 classes). This means that the hard-class maps are more reliable when aggregating classes to a higher level in the hierarchy (e.g. 'Discontinuous Urban Fabric' and 'Continuous Urban Fabric' to 'Urban Fabric'). Some single-class probabilities may more closely represent actual patterns for some classes that were overshadowed by unequal sample point distributions. Users are encouraged to set their own thresholds when postprocessing these datasets to optimize the accuracy for their specific use case. Uncertainty quantification: Uncertainty is quantified by taking the standard deviation of the probabilities predicted by the three components of the spatiotemporal ensemble model. Data validation approaches: The LULC classification was validated through spatial 5-fold cross-validation as detailed in the accompanying publication. Completeness: The dataset has chunks of empty predictions in regions with complex coast lines (e.g. the Zeeland province in the Netherlands and the Mar da Palha bay area in Portugal). These are artifacts that will be avoided in subsequent versions of the LULC product. Consistency: The accuracy of the predictions was compared per year and per 30km*30km tile across europe to derive temporal and spatial consistency by calculating the standard deviation. The standard deviation of annual weighted F1-score was 0.135, while the standard deviation of weighted F1-score per tile was 0.150. This means the dataset is more consistent through time than through space: Predictions are notably less accurate along the Mediterrranean coast. The accompanying publication contains additional information and visualisations. Positional accuracy: The raster layers have a resolution of 30m, identical to that of the Landsat data cube used as input features for the machine learning framework that predicted it. Temporal accuracy: The dataset contains predictions and uncertainty layers for each year between 2000 and 2019. Thematic accuracy: The maps reproduce the Corine Land Cover classification system, a hierarchical legend that consists of 5 classes at the highest level, 14 classes at the second level, and 44 classes at the third level. Class 523: Oceans was omitted due to computational constraints.