Regional model ICON-D2 The DWD's ICON-D2 model is a forecast model which is operated for the very-short range up to +27 hours (+45 hours for the 03 UTC run). Due to its fine mesh size, the ICON-D2 especially provides for improved forecasts of hazardous weather conditions, e.g. weather situations with high-level moisture convection (super and multi-cell thunderstorms, squall lines, mesoscale convective complexes) and weather events that are influenced by fine-scale topographic effects (ground fog, Föhn winds, intense downslope winds, flash floods). The model area of ICON-D2 covers the whole German territory, Benelux, Switzerland, Austria and parts of the other neighbouring countries at a horizontal resolution of 2.2 km. In the vertical, the model defines 65 atmosphere levels. The fairly short forecast periods make perfect sense because of the purpose of ICON-D2 (and its small model area). Based on model runs at 00, 06, 09, 12, 15, 18 and 21 UTC, ICON-D2 provides new 27-hour forecasts every 3 hours. The model run at 03 UTC even covers a forecast period of 45 hours. The ICON-D2 forecast data for each weather element are made available in standard packages at our free DWD Open Data Server, both on a rotated grid and on a regular grid. Regional ensemble forecast model ICON-D2 EPS The ensemble forecasting system ICON-D2 EPS is based on the DWD's numerical weather forecast model ICON-D2 and currently includes 20 ensemble members. All ensemble members are calculated at the same horizontal grid spacing as the operational configuration of ICON-D2 (2.2 km). Like ICON-D2, the ICON-D2 EPS ensemble system provides forecasts up to +27 hours for the same model area (up to +45 hours based on the 03 UTC run). For generating the ensemble members, some of the features of the forecasting system are changed. The method currently used to generate the ensemble members involves varying the - lateral boundary conditions - initial state - soil moisture - and model physics. For varying the lateral boundary conditions and the initial state, forecasts from various global models are used. The ICON-D2 EPS is provided on the DWD Open Data Server in the native triangular grid. Note: All previously COSMO-D2 based aviation weather products have been migrated to ICON-D2 on 10.02.2021. However, the familiar design of these products remains unchanged.

Overview: osm: Military rasterized from OSM landuse polygons, first to 10m spatial resolution and after downsampled to 30m by spatial average. Traceability (lineage): The class-wise layers of this dataset were extracted from OpenStreetMap data downloaded from geofabrik.de and aggregated based on labels assigned to the volunteered geographical information objects. Scientific methodology: nan Usability: The extracted classes can be used to preprocess training data (as detailed in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ). Users are advised to remember the potential inconsistencies in volunteered geographical information, however: Some regions of Europe have been less consistently mapped in OpenStreetMap. This may introduce bias in any subsequent modelling. Uncertainty quantification: nan Data validation approaches: This dataset has not been validated Completeness: Volunteered geographical information often more complete in regions with more active contributors. It is likely that this dataset contains many omission errors in regions of Europe where OpenStreetMap is used less intensively. Consistency: Volunteered geographical information often more complete in regions with more active contributors. It is likely that this dataset contains many omission errors in regions of Europe where OpenStreetMap is used less intensively. Positional accuracy: The rasters have a spatial resolution of 30m Temporal accuracy: The maps are based on an extract from 2020. Thematic accuracy: The 30m pixels of each OSM extract map have values ranging from 0-100, indicating the density aggregated from 10m pixels where rasterized objects burned the value 100 in a 0-value raster.

Overview: 142: Areas used for sports, leisure and recreation purposes. 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.

Overview: osm: Meadows rasterized from OSM landuse polygons, first to 10m spatial resolution and after downsampled to 30m by spatial average. Traceability (lineage): The class-wise layers of this dataset were extracted from OpenStreetMap data downloaded from geofabrik.de and aggregated based on labels assigned to the volunteered geographical information objects. Scientific methodology: nan Usability: The extracted classes can be used to preprocess training data (as detailed in Witjes et al., 2022 (in review, preprint available at https://doi.org/10.21203/rs.3.rs-561383/v3 ). Users are advised to remember the potential inconsistencies in volunteered geographical information, however: Some regions of Europe have been less consistently mapped in OpenStreetMap. This may introduce bias in any subsequent modelling. Uncertainty quantification: nan Data validation approaches: This dataset has not been validated Completeness: Volunteered geographical information often more complete in regions with more active contributors. It is likely that this dataset contains many omission errors in regions of Europe where OpenStreetMap is used less intensively. Consistency: Volunteered geographical information often more complete in regions with more active contributors. It is likely that this dataset contains many omission errors in regions of Europe where OpenStreetMap is used less intensively. Positional accuracy: The rasters have a spatial resolution of 30m Temporal accuracy: The maps are based on an extract from 2020. Thematic accuracy: The 30m pixels of each OSM extract map have values ranging from 0-100, indicating the density aggregated from 10m pixels where rasterized objects burned the value 100 in a 0-value raster.

Overview: 211: Slope of non-irrigated arable land derived by OLS regression over the probabilities values (2000—2019). The std. error of the model was considered as uncertainty. 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.

Overview: 244: Annual crops or grazing land under the wooded cover of forestry species. 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.

311: R2 of OLS regression calculated over the probabilities values (2000—2019) for broad-leaved forest.

421: Vegetated low-lying areas in the coastal zone, above the high-tide line, susceptible to flooding by seawater. Often in the process of being filled in by coastal mud and sand sediments, gradually being colonized by halophilic plants. Salt marshes are in most cases directly connected to intertidal areas and may successively develop from them in the long-term. Salt-pans for extraction of salt from salt water by evaporation, active or in process of abandonment. Sections of salt marsh exploited for the production of salt, clearly distinguishable from the rest of the marsh by their parcellation and embankment systems. Coastal zone under tidal influence between open sea and land, which is flooded by sea water regularly twice a day in a ca. 12 hours cycle. Area between the average lowest and highest sea water level at low tide and high tide. Generally non-vegetated expanses of mud, sand or rock lying between high and low water marks. The seaward boundary of intertidal flats may underlay constant change in geographical extent due to littoral morphodynamics. Range of water level between low tide and high tide may vary between decimeters and several meters in height.

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

Overview: 312: Vegetation formation composed principally of trees, including shrub and bush understorey, where coniferous species predominate. The predominant classifying parameter for this class is a crown cover density of > 30 % or a minimum 500 subjects/ha density, with coniferous trees representing > 75 % of the formation. The minimum tree height is 5 m (with the exception of Christmas tree plantations). 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.