Data in Brief 60 (2025) 111613 Contents lists available at ScienceDirect Data in Brief journal homepage: www.elsevier.com/locate/dib Data Article Pan-European forest maps produced with a combination of earth observation data and national forest inventory plots Jukka Miettinen a , ∗, Johannes Breidenbach b , ∗, Patricia Adame c , Radim Adolt d , Iciar Alberdi c , Oleg Antropov a , Ólafur Arnarsson e , Rasmus Astrup b , Ambros Berger f , Jón Bogason e , Gherardo Chirici g , Piermaria Corona h , Giovanni D’Amico g , Jiří Fejfar d , Christoph Fischer i , Florence Gohon j , Thomas Gschwantner f , Johannes Hertzler k , Zsofia Koma b , Kari T. Korhonen l , Luka Krajnc m , Nicolas Latte n , Philippe Lejeune n , Andrew McCullagh o , Marcin Mionskowski p , Daniel Moreno-Fernández c , Mari Myllymäki l , Mats Nilsson q , Jérôme Perin n , Juho Pitkänen l , John Redmond o , Thomas Riedel k , Johannes Schumacher b , Lauri Seitsonen a , Laura Sirro a , Mitja Skudnik m , r , Arnór Snorrason e , Radosław Sroga p , Berthold Traub i , Björn Traustason e , Bertil Westerlund q , Stephanie Wurpillot j a VTT Technical Research Centre of Finland, VTT, P.O. Box 10 0 0, FI-02044, Finland b NIBIO Norwegian Institute of Bioeconomy Research, Division of Forest and Forest Resources, Høgskoleveien 8, 1430 Ås, Norway c Institute of Forest Sciences (ICIFOR, INIA-CSIC), Crta. de A Coruña km 7.5, Madrid, E-28040, Spain d Forest Management Institute (UHUL), Nábřežní 1326, Brandýs nad Labem, 250 01, Czech Republic e Land and Forest Iceland, Mógilsá, Reykjavík, 162, Iceland f Austrian Research Centre for Forests (BFW), Seckendorff-Gudent-Weg 8, Vienna, 1131, Austria g Dipartimento di Scienze e Tecnologie Agrarie, Università degli Studi di Firenze, Alimentari, Ambientali e Forestali. Via San Bonaventura 13, Firenze, 50145, Italy h CREA Research Centre for Forestry and Wood, viale Santa Margherita 80, Arezzo, 52100, Italy i Swiss Federal Research Institute WSL, Zürcherstrasse 111, Birmensdorf, CH 8903, Switzerland j Institut National de l’information Géographique et Forestière (IGN), Service de l’information Statistique Forestière et Environnementale, Chemin du Château des Barres, Nogent-Sur-Vernisson, 45290, France k Thünen Institute of Forest Ecosystems, Alfred-Möller-Straße 1, Haus 41/42, Eberswalde, 16225, Germany l Natural Resources Institute Finland (Luke), P.O. Box 2, Helsinki, FI-00791, Finland m Slovenian Forestry Institute, Večna pot 2, Ljubljana, 10 0 0, Slovenia ∗ Corresponding authors. E-mail addresses: Jukka.Miettinen@vtt.fi (J. Miettinen), Johannes.Breidenbach@nibio.no (J. Breidenbach). https://doi.org/10.1016/j.dib.2025.111613 2352-3409/© 2025 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ) https://doi.org/10.1016/j.dib.2025.111613 http://www.ScienceDirect.com/science/journal/23523409 http://www.elsevier.com/locate/dib http://crossmark.crossref.org/dialog/?doi=10.1016/j.dib.2025.111613&domain=pdf mailto:Jukka.Miettinen@vtt.fi mailto:Johannes.Breidenbach@nibio.no https://doi.org/10.1016/j.dib.2025.111613 http://creativecommons.org/licenses/by/4.0/ 2 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 n Gembloux Agro-Bio Tech (GxABT), Forest Resource Management Unit, University of Liège (ULiège), 2 Passage des Déportés, Gembloux, 5030, Belgium o Forest Service, Department of Agriculture, Food and the Marine, Agriculture House, Kildare Street, Dublin 2, Ireland p Bureau for Forest Management and Geodesy, Leśników 21, Raszyn, 05-090, Poland q Department of Forest Resource Management, Swedish University of Agricultural Sciences (SLU), Skogsmarksgränd, Umeå, SE-901 83, Sweden r Biotechnical Faculty, University of Ljubljana, Jamnikarjeva 101, Ljubljana, 10 0 0, Slovenia a r t i c l e i n f o Article history: Received 28 November 2024 Revised 7 April 2025 Accepted 28 April 2025 Available online 6 May 2025 Dataset link: High-Resolution Pan-European Forest Structure Maps: An Integration of Earth Observation and National Forest Inventory Data (Original data) Keywords: European forest monitoring system Remote sensing In-situ data Forest attribute maps a b s t r a c t The dataset includes Pan-European maps of timber vol- ume (Vol), above-ground biomass (AGB), and deciduous- coniferous proportion (DCP) with a pixel size of 10 ×10 m for the reference year 2020. In addition, a measure of prediction uncertainty is provided for each pixel. The maps have been created using a combination of a Sentinel-2 mosaic, Coperni- cus layers, and National Forest Inventory (NFI) data. The mapping was done with the k-Nearest Neighbour (kNN, k = 7) approach with harmonized data of species-specific Vol and AGB from 14 NFIs consisting of approximately 151 0 0 0 field plots across Europe. The maps cover 40 European coun- tries, 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 % in the South-Eastern area. The maps are on average nearly unbiased on European level (1.0 % of the mean AGB), but show significant over- estimation for small biomass values (53 % bias for forests with AGB less than 150 t/ha) and underestimation for high biomass values (-55 % bias for forests with AGB higher than 500 t/ha). The created maps are the first of their kind as they are uti- lizing a large number of harmonized NFI plot observations and consistent remote sensing data for high-resolution forest attribute mapping. While the published maps can be use- ful 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 proce- dures were not applied, and especially high Vol and AGB val- ues tend to be underestimated. We therefore discourage from summarizing map values (pixel counting) over areas in inter- est, as this may inadvertently result in biased estimates. © 2025 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ) https://zenodo.org/records/13143235 http://creativecommons.org/licenses/by/4.0/ J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 3 Specifications Table Subject Earth & Environmental Sciences Specific subject area Pan-European maps of timber volume (Vol), above-ground biomass (AGB), and deciduous-coniferous proportion (DCP) with a pixel size of 10 ×10 m for year 2020. Type of data Uint16 GeoTiff files provided in 50 0 ×50 0 km tiles in the EPSG:3035 - ETRS89-extended / LAEA Europe projection. Data collection The maps were produced with the k-Nearest Neighbour (kNN) approach (parameters: k = 7, Euclidian distance, weighing inverse to distance). Input data consisted of a harmonized database of species-specific Vol and AGB from 14 National Forest Inventories with around 151 0 0 0 field plots across Europe, seven spectral bands of Sentinel-2 satellite (B2, B3, B4, B5, B8, B11, B12) and two auxiliary layers (Copernicus high resolution forest type and tree cover density). The processing was run on the Forestry TEP platform ( https://f-tep.com/ ). Data source location The maps cover 40 European countries, forming a continuous coverage of the western part of the European continent. Data accessibility Repository name: Zenodo Data identification number: 10.5281/zenodo.13143235 Direct URL to data: 10.5281/zenodo.13143235 Instructions for accessing these data: Download requires no registration or any other procedure Related research article None. 1. Value of the Data • The maps allow a visualization of the spatial distribution of key forest attributes across Eu- rope. • The maps will reduce the uncertainty of estimates when combined with reliable reference data in the model-assisted framework. • The maps can be the starting point for forecasting models for predicting future forest devel- opment that require high-resolution input data. • The maps provide a base for decision support in regions where no other forest information is available. • The maps give an impression of the local variation in forest attributes that can be used as pre-information for more in-depth surveys. 2. Background The first European wide growing stock volume and biomass maps combining NFI data with remote sensing datasets were created with 50 0-10 0 0 m resolution NOAA-AVHRR and MODIS data [ 1 , 2 ]. Since then, several remote sensing based global products covering Europe have been produced using both optical and radar satellite sensors [ 3–5 ] reaching as high as 100 m spatial resolution. Although these types of maps are valuable for many aspects of forest monitoring, relying solely on summarizing remote sensing-based maps over large regions such as entire countries, can lead to significant systematic errors. Therefore, the EU PathFinder project [ 6 ] aims to de- velop and demonstrate a forest monitoring system that allows reporting to European and global policies. A mapping and estimation system integrating the use of remote sensing and field data aims to facilitate effective use of field data with remotely sensed and other auxiliary datasets to produce precise forest information. The maps published here are the first maps produced with the system using a combination of European wide Sentinel-2 imagery and over 150 0 0 0 NFI field plots across Europe. https://f-tep.com/ https://doi.org/10.5281/zenodo.13143235 http://10.5281/zenodo.13143235 4 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 Fig. 1. European wide AGB (t/ha) and DCP (%) maps with subsets showing fine details. 3 a m p d t T ( . Data Description Maps of three different tar get variables are made available, including timber volume (Vol), bove-ground biomass (AGB), and deciduous-coniferous proportion (DCP) [ 7 ]. While the DCP ap provides the percentage of conifers, this allows straightforward calculation of the broadleaf roportion as 100-DCP. Fig. 1 illustrates the AGB map with subsets of AGB and DCP maps from ifferent parts of Europe. In addition to the target variable maps, also standard deviation layers are provided for each arget variable. All of the maps were masked with the Copernicus High Resolution Layer Forest ype (FTY) 2018 forest extent. Areas outside forest have been masked as non-forest and no-data Table 1 ). Table 1 Output map technical characteristics. Variable Variable naming Pixel values Format Volume Vol (attribute) stdev_vol (uncertainty) 65535: No data 65534: Non-forest Other values: m ³/ha 10 m UInt16 GeoTiff Above Ground Biomass AGB (attribute) stdev_agb (uncertainty) 65535: No data 65534: Non-forest Other values: t/ha 10 m UInt16 GeoTiff Conifer proportion P_agb_conifers (attribute) stdev_P_agb_conifers (uncertainty) 65535: No data 65534: Non-forest Other values: % 10 m UInt16 GeoTiff J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 5 Fig. 2. Maps are provided in 10 m spatial resolution in 50 0 ×50 0 km tiles in the EPSG:3035 - ETRS89-extended / LAEA Europe projection. Volume map as background in the image. All maps are provided in 10 m spatial resolution in 50 0 ×50 0 km tiles in the EPSG:3035 - ETRS89-extended / LAEA Europe projection ( Fig. 2 ). The file naming follows the following pat- tern: ‘Year’_’variable’_’tile’.tif, with variable names as defined in Table 1 . For example, the standard deviation map of volume for the 50 0 ×50 0 km grid tile with the lower left corner E 3 900 000 m and N 2 400 000 m is ‘2020_stdev_vol_E39_N24.tif’. The maps fully cover a total of 40 European countries. In addition to the 27 EU countries, this includes Albania, Andorra, Bosnia and Herzegovina, Holy See, Liechtenstein, Monaco, Mon- tenegro, North Macedonia, Norway, San Marino, Serbia, Switzerland and United Kingdom. The maps were produced using 13 separate processing areas (see the next section for more details). Error metrics for each of the processing areas are provided in Table 2 and Fig. 3 . Error metrics have been calculated by using the validation plots. For the processing areas where NFI plots are available, this approach can be expected to produce reliable error metrics. However, for the processing areas that do not have plots (i.e. Areas 2, 3, 5, 7, 8, 9 and 13), the error metrics should be treated with caution. For these areas, the mapping and validation plots 6 J. M iettin en , J. B reid en b a ch a n d P. A d a m e et a l. / D a ta in B rief 6 0 (2 0 2 5 ) 1116 13 Table 2 Plot level error metrics for the 13 processing areas calculated from the validation plots. Areas where plots were available and the plot location was used in the selection of the neighbours are in bold , whereas areas where plots were not available are in italic . Processing area Vol (m3 /ha) AGB (t/ha) Con% (%) RMSE RMSE% Bias Bias% R2 RMSE RMSE% Bias Bias% R2 RMSE RMSE% Bias Bias% R2 Area 1 75.7 58.2 -0.8 -0.6 0.54 40.2 53.2 -0.7 -0.9 0.54 22.3 30.4 1.1 1.5 0.52 Area 2 124.5 57.8 0.3 0.1 0.38 70.0 55.4 0.0 0.0 0.39 22.8 33.7 0.6 0.9 0.61 Area 3 187.7 55.4 0.7 0.1 0.05 103.4 58.0 0.5 0.3 0.08 22.6 37.7 -0.2 -0.3 0.73 Area 4 129.8 64.8 -0.5 -0.3 0.53 77.8 54.0 -1.0 -0.7 0.48 28.4 42.1 0.1 0.1 0.58 Area 5 146.4 72.0 9.3 4.6 0.38 90.6 63.4 4.3 3.0 0.31 27.0 50.6 0.7 1.4 0.67 Area 6 177.4 56.6 8.2 2.6 0.29 107.5 56.6 4.8 2.5 0.24 22.3 36.8 0.3 0.5 0.72 Area 7 201.0 61.9 7.9 2.4 0.27 127.2 63.4 5.3 2.6 0.15 24.8 41.3 -0.4 -0.7 0.65 Area 8 181.3 73.6 6.0 2.4 0.37 110.4 72.0 2.7 1.8 0.25 24.9 59.1 -0.1 -0.2 0.69 Area 9 127.0 76.9 2.3 1.4 0.41 86.1 73.4 3.3 2.8 0.30 19.1 110.3 -0.9 -5.1 0.69 Area 10 168.6 61.7 1.7 0.6 0.30 98.1 59.5 0.8 0.5 0.23 24.1 71.6 0.1 0.3 0.68 Area 11 178.2 64.3 5.2 1.9 0.31 103.6 62.3 2.7 1.6 0.24 24.2 61.8 0.1 0.3 0.69 Area 12 115.0 74.6 -0.9 -0.6 0.48 74.4 67.6 -0.6 -0.5 0.41 25.2 64.8 -0.3 -0.8 0.69 Area 13 49.5 83.9 -2.1 -3.5 0.60 44.6 71.6 -1.0 -1.6 0.47 26.5 65.5 -0.5 -1.2 0.68 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 7 Fig. 3. Relative plot level accuracies of AGB by processing areas. were manually compiled using plots from ecologically similar regions. Although the selection was made with best available knowledge of the areas, the error metrics for the areas without field plots may be overly optimistic. Scatterplots based on the validation data show that the bulk of the data follow the 1:1 line. However, high values tend to be underestimated and low values tend to be overestimated ( Fig. 4 ). These tendencies are evident in the change of bias for different size classes of forest. While the maps are on average nearly unbiased on European level (1.4 t/ha or 1.0 % of the 137 t/ha mean AGB), there is a clear overestimation for small biomass classes and increasing under- estimation with increasing biomass ( Table 3 ). Table 3 Bias of AGB values for different biomass classes. AGB bin (t/ha) 0-124 125-249 250-374 375-499 500- > Bias (t/ha) 38.3 -3.5 -87.1 -174.0 -351.1 Bias% 66.3 -1.9 -29.1 -42.6 -58.7 8 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 Fig. 4. Scatterplots of observed and predicted Vol, AGB, and DCP based on the validation data. Darker colors indicate a higher density of observations. 4 4 l t m 4 A v t t s . Experimental Design, Materials and Methods .1. NFI Field Data Processing The workflow chart of the data processing is illustrated in Fig. 5 . The plot level data calcu- ations of harmonized forest structure variables and the extraction of Sentinel-2 spectral signa- ures for plots were conducted by national NFI organizations. Detailed definitions of the har- onized volume (V) and above-ground biomass (AGB) can be found at [ 8 ] (variable ids 2 and ). The deciduous-coniferous proportion (DCP) was calculated on plot level as the proportion of GB of coniferous species [ 9 ] at the plot. The deciduous proportion is the inverse of the DCP alue. As a rule, plots measured 2019-2021 were used, to minimize the temporal difference between he field measurement and the Sentinel-2 observation (2020). However, in some countries the emporal range was extended either to include a sufficient number of plots or to include mea- urements from geographic areas that otherwise would not have been covered ( Table 4 ). Because J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 9 Fig. 5. Workflow chart of data processing. Original data sources highlighted in yellow, intermediate in orange and the final output data in green. Further details of the datasets and processing steps can be found in the text. Table 4 Overview of the NFI data used in the mapping. Country code Measurement years Number of plots used for mapping Number of plots used for validation Plots/forest area (N/km2 ) AT 2019-2021 3816 1908 0.14 BE 2019-2021 934 466 0.29 CH 2019-2022 2168 1083 0.26 CZ 2018-2020 6902 3450 0.36 DE 2017 5735 2867 0.08 ES 2018-2021 12592 6296 0.10 FI 2019-2021 21432 10716 0.16 FR 2019-2021 12390 6195 0.11 IE 2020-2022 1185 592 0.24 IT 2018-2019 2813 1406 0.05 NO 2019-2021 5908 2954 0.07 PL 2019-2021 15754 7877 0.26 SE 2019-2021 8602 4300 0.06 SI 2018 566 283 0.07 TOTAL - 100797 50393 the Sentinel-2 and Copernicus data are organized in overlapping tiles, some sample plots were used more than once for modelling. The spatial distribution of the plots is shown in Fig. 6 . All European NFIs used in the creation of the maps use representative probability sampling designs to select field sample plots that are used to create official statistics of forest resources. More detailed description of the NFI designs can be found e.g. in [ 10 , 11 ]. 10 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 Fig. 6. Number of field plots within Sentinel-2 tiles in countries that provided NFI data. 4 n C m a p a .2. Remote Sensing and Auxiliary Data Processing The following remotely sensed data were used in the production of the maps ( Fig. 5 ): • A European-wide Sentinel-2 mosaic of composite images for the year 2020 [ 12 ] was used as the primary remote sensing data source. The image compositing algorithm using Level 2A surface reflectance products is described in Miettinen et al. [ 13 ]. The final composite images included seven spectral bands (B2, B3, B4, B5, B8, B11 and B12), all resampled into 10 m spatial resolution. • Copernicus High Resolution Layers for forest in 10 m resolution with reference year 2018. We used Forest Type (FTY; coniferous forest, broadleaved forest, non-forest) and Tree Cover Density (TCD; 0-100 %) [ 14 ]. • Global Forest Change (GFC) 30 m resolution forest canopy loss product [ 15 ]. We used version 1.9 that included the years of forest canopy losses between and including 2001 and 2021. The Sentinel-2 and Copernicus products created the feature space when choosing the nearest eighbours in the k Nearest Neighbour mapping (see more details below). The Global Forest hange product was used only in screening plots that had experienced changes after the field easurements. The Sentinel-2 bands and the TCD data had continuous values and the weighted mean for circle with a size of 100 m2 centred on each NFI sample plot was calculated from them. The ixel proportions covering the circle were used as weights. For the categorical variables (FTY nd GFC), the weighted mode was calculated for the same circle. The weighted mode was the J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 11 Fig. 7. Sentinel-2 tiles colorized by processing area. Left: processing areas where NFI plots were available and location information was used in modelling. Right: processing areas where NFI plots were not available and location information was not used in modelling (except for northing in Area 2). Plot sourcing areas are shown in transparent color. category with the greatest sum of weights. The calculations were done using R by each NFI organization, in order to use the exact plot coordinates. The standardized R scripts for this task are available as a git repository [ 16 ]. 4.3. Application of the kNN Approach The k Nearest Neighbour (kNN) approach was used with training data including forest target variables (V, AGB, and DCP) and the corresponding remotely sensed variables (seven Sentinel-2 bands, TCD, and FTY) for each NFI plot. Furthermore, 1 km INSPIRE grid cell locations (northing and easting) were included in the feature space to enable utilization of the plot location when selecting the nearest neighbours. The kNN model was run with the same parameters (k = 7, Eu- clidian distance, weighing neighbours inverse to distance) in all processing areas. All features used in the modelling had equal weights. The data were screened to (1) exclude cloud contaminated observations (based on compos- ite image quality band accepting only values greater than 40 0 0) and (2) exclude plots where changes had been detected by GFC since 2018 (or a year before the earliest plot measurements). The screened data were sorted by volume and every third plot was assigned to the validation data set. The remaining two thirds of the plots were used for the mapping ( Table 4 ). The same dataset was used to predict all the three variables simultaneously in a multivariate manner. For the production of the maps, 13 processing areas were created taking into account the ge- ographical areas of Europe and the availability of field sample plots ( Fig. 7 ). Six of the processing areas contained NFI plots, while the remaining processing areas did not have any NFI plots. For those areas without NFI plots, training data were created using plots from ecologically similar sourcing areas. Mean Vol ranged between 130 and 326 m3 /ha in the processing areas, whereas mean AGB values ranged between 76 and 202 t/ha. The proportion of conifer-dominated plots ranged from more than 70 % to less than 40 %. In regions with NFI plots, the processing area borders were located in the middle of countries, but all plots from the countries that were (even partially) covered by a processing area were used for mapping. This approach ensured that in adjacent processing areas plots from both sides of the processing area border were used, resulting in a smooth transition between the processing areas with no visible changes in the map attributes at the borders of the processing areas. For areas where NFI plots were available, the INSPIRE 1 km grid northing and easting were used as features in the selection of the nearest neighbours. This approach allowed utilization of 12 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 l i s S b S p w l c u T w o v r c o ( q m t m c l t b N b arge training datasets while still ensuring that the plots used for mapping were from ecolog- cally similar areas. For countries where NFI plots were not available, training data were con- tructed with nearest available NFI plots from neighbouring countries. In these cases, the IN- PIRE 1 km grid locations were not used among the attributes to search for the closest neigh- ours. An exception was processing area 2 ( Fig. 7 ), where northing was used. All processing was conducted in the Forestry TEP [ 17 ] by Sentinel-2 tiles. A total of 745 entinel-2 tiles were processed. Euclidean distance was used to select the nearest neighbours in the feature space. The k-NN rediction ˆ yp for pixel p is given by ˆ yp = ∑ l wl yl (1) here yl is the vector of observations and wl is the weight of the l’th nearest neighbor with = 1.….k. Weights inverse to the Euclidean distance were used. In addition to the prediction, we alculated the standard deviation ˆ sp of the nearest neighbors for each pixel as a measure of ncertainty ˆ sp = √ ∑ l ( yl − ˆ yp )2 k (2) Root-mean-square-error (RMSE), bias and R2 were used to evaluate the mapping results. hese metrics were calculated by predicting the response variables for the validation plots. RMSE = √ ∑ i ( yi − ̂ yi ) 2 n (3) Bias = ∑ i ( yi − ̂ yi ) n (4) R2 = 1 − ∑ i ( yi − ̂ yi ) 2 ∑ i ( yi − yi ) 2 (5) here yi represents the reference values, ˆ yi represents the predicted values, i = 1.…. n indexes the bservations, and n is the number of observations in the validation database. The RMSE and bias alues were also compared to the mean value of the variable in the validation plots, deriving elative metrics, denoted as RMSE% and Bias%. The above error metrics were calculated by using the validation plots from the same pro- essing areas, where available. To investigate the potential effects of utilizing data gathered from ther geographical areas to derive error metrics, predictions were made for Czechia and Poland which are both included in processing Area 6), without using the NFI data from the country in uestion. The results were validated with the validation data from the country. The error metrics were calculated using (1) Area 6 data, and (2) Area 6 data excluding the apping plots from the respective country ( Table 4 ). In both cases. validation plots only from he respective country were used. In case 2), location information was not used in the model, aking it comparable to those processing areas that do not have NFI plots available. It can clearly be seen that in both cases the predictions with the national NFI plots produce learly lower relative RMSE values ( Table 5 ). This highlights the importance of the availability of ocal reference data when using empirical methods such as the k-NN approach. The biases, on he other hand, behave more erratically and vary strongly between the training dataset and can e high in some cases. Overall, the target variable predictions for the processing areas without FI plots need to be treated with caution. There is an increased probability of high biases caused y the use of NFI plots from other geographical areas. J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 13 Table 5 Error metrics with and without NFI plots for two example countries. Country Training dataset Vol AGB Con% RMSE% Bias% RMSE% Bias% RMSE% Bias% Czechia (CZ) Current map 53.8 2.3 56.9 -0.6 35.7 -0.3 No CZ plots 63.2 1.2 66.0 -15.7 39.8 0.6 Poland (PL) Current map 54.6 3.1 53.1 3.4 35.4 0.7 No PL plots 65.8 22.1 65.6 23.1 39.1 4.4 Limitations It is important to note the underestimation of high and overestimation of low volume and biomass values in the maps ( Fig. 4 , Table 3 ). It is therefore not recommended to use the maps as a sole means for estimation of forest statistics for a given interest area. The maps are primarily meant to be used in model-assisted estimation where model-related biases can be mitigated, and field-based estimates improved. In addition to the uncertainty of the accuracy of the maps in areas that have no field plots (discussed in the previous section), another issue potentially affecting the consistency of the map is the difference in the NFI designs. That means that, e.g. the temporal and spatial fre- quency of reference plots differ between the countries. In some countries, the plots used for the creation of the maps were measured a few years before the satellite data acquisition, while in most countries the temporal match is very good ( Table 4 ). The effects of these differences could not be evaluated for the current set of maps but should be a topic for further investigation. Ethics Statement The authors confirm that they have read and follow the ethical requirements for publication in Data in Brief and confirm that the current work does not involve human subjects, animal experiments, or any data collected from social media platforms. Data Availability High-Resolution Pan-European Forest Structure Maps: An Integration of Earth Observation and National Forest Inventory Data (Original data) (Zenodo). CRediT Author Statement Jukka Miettinen: Conceptualization, Methodology, Formal analysis, Writing – original draft, Writing – review & editing; Johannes Breidenbach: Conceptualization, Methodology, Writing – original draft, Writing – review & editing, Funding acquisition, Project administration; Patricia Adame: Resources, Data curation; Radim Adolt: Resources, Data curation; Iciar Alberdi: Re- sources, Data curation; Oleg Antropov: Methodology, Software; Ólafur Arnarsson: Resources, Data curation; Rasmus Astrup: Project administration, Writing – review & editing; Ambros Berger: Methodology, Resources, Data curation, Writing – review & editing; Jón Bogason: Re- sources, Data curation, Writing – review & editing; Gherardo Chirici: Resources, Data curation; Piermaria Corona: Resources, Data curation; Giovanni D’Amico: Resources, Data curation, Writ- ing – review & editing; Jiří Fejfar: Resources, Data curation; Christoph Fischer: Resources, Data curation; Florence Gohon: Resources, Data curation; Thomas Gschwantner: Resources, Data cu- ration, Writing – review & editing; Johannes Hertzler: Resources, Data curation; Zsofia Koma: https://zenodo.org/records/13143235 14 J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 P t L s i i P D S M s D c A D t R [ [ roject administration, Writing – review & editing; Kari T. Korhonen: Resources, Data cura- ion; Luka Krajnc: Resources, Data curation; Nicolas Latte: Resources, Data curation; Philippe ejeune: Resources, Data curation; Andrew McCullagh: Resources, Data curation; Marcin Mion- kowski: Resources, Data curation; Daniel Moreno-Fernández: Resources, Data curation, Writ- ng – review & editing; Mari Myllymäki: Methodology, Data curation, Writing – review & edit- ng; Mats Nilsson: Resources, Data curation; Jérôme Perin: Resources, Data curation; Juho itkänen: Resources, Data curation, Writing – review & editing; John Redmond: Resources, ata curation; Thomas Riedel: Resources, Data curation, Writing – review & editing; Johannes chumacher: Resources, Data curation; Lauri Seitsonen: Software, Visualization; Laura Sirro: ethodology, Formal analysis; Mitja Skudnik: Resources, Data curation; Arnór Snorrason: Re- ources, Data curation; Radosław Sroga: Resources, Data curation; Berthold Traub: Resources, ata curation; Björn Traustason: Resources, Data curation; Bertil Westerlund: Resources, Data uration; Stephanie Wurpillot: Resources, Data curation. cknowledgements This study received funding from the EU under GA101056907 ( PathFinder). eclaration of Competing Interest The authors declare that they have no known competing financial interests or personal rela- ionships that could have appeared to influence the work reported in this paper eferences [1] R. Päivinen, J. Van Brusselen, A. Schuck, The growing stock of European forests using remote sensing and forest inventory data, For. Int. J. For. Res. 82 (2009) 479–490, doi: 10.1093/forestry/cpp017 . [2] H. Gallaun, G. Zanchi, G.-J. Nabuurs, G. Hengeveld, M. Schardt, P.J. Verkerk, EU-wide maps of growing stock and above-ground biomass in forests based on remote sensing and field measurements, For. Ecol. Manage 260 (2010) 252–261, doi: 10.1016/j.foreco.2009.10.011 . [3] M. Thurner, C. Beer, M. Santoro, N. Carvalhais, T. Wutzler, D. Schepaschenko, A. Shvidenko, E. Kompter, B. Ahrens, S.R. Levick, C. Schmullius, Carbon stock and density of northern boreal and temperate forests, Glob. Ecol. Biogeogr. 23 (2014) 297–310, doi: 10.1111/geb.12125 . [4] T. Hu, Y. Su, B. Xue, J. Liu, X. Zhao, J. Fang, Q. Guo, Mapping global forest aboveground biomass with spaceborne LiDAR, optical imagery, and forest inventory data, Remote Sens. (Basel) 8 (2016) 565, doi: 10.3390/rs8070565 . [5] M. Santoro, O. Cartus, N. Carvalhais, D.M.A. Rozendaal, V. Avitabile, A. Araza, S. de Bruin, M. Herold, S. Que- gan, P. Rodríguez-Veiga, H. Balzter, J. Carreiras, D. Schepaschenko, M. Korets, M. Shimada, T. Itoh, Á.M. Martínez, J. Cavlovic, R. Cazzolla Gatti, P. da Conceição Bispo, N. Dewnath, N. Labrière, J. Liang, J. Lindsell, E.T.A. Mitchard, A . Morel, A .M. Pacheco Pascagaza, C.M. Ryan, F. Slik, G. Vaglio Laurin, H. Verbeeck, A. Wijaya, S. Willcock, The global forest above-ground biomass pool for 2010 estimated from high-resolution satellite observations, Earth. Syst. Sci. Data 13 (2021) 3927–3950, doi: 10.5194/essd- 13- 3927- 2021 . [6] PathFinder, Towards an integrated consistent European LULUCF monitoring and policy pathway assessment frame- work. https://pathfinder-heu.eu/ , 2024 (accessed 19 November 2024). [7] J. Miettinen, P. Adame, R. Adolt, I. Alberdi, O. Antropov, Ó. Arnarsson, R. Astrup, A. Berger, J. Bogason, G. Chirici, P. Corona, G. D’Amico, C. Fischer, F. Gohon, T. Gschwantner, J. Hertzler, Z. Koma, K.T. Korhonen, L. Krajnc, N. Latte, P. Lejeune, A. McCullagh, M. Mionskowski, D. Moreno, M. Myllymäki, M. Nilsson, J. Perin, J. Pitkänen, J. Redmond, T. Riedel, J. Schumacher, L. Seitsonen, L. Sirro, M. Skudnik, A. Snorrason, R. Sroga, B. Traub, S. Wurpillot, J. Breiden- bach, High-resolution pan-European forest structure maps: an integration of earth observation and national forest inventory data, Zenodo V1 (2024), doi: 10.5281/zenodo.13143235 . [8] nFIESTA, PathFinder demo study, codelist for target variables. https://gitlab.com/nfiesta/pathfinder_demo_study/-/ wikis/Codelist%20of%20target%20variables , 2023 (accessed 19 November 2024). [9] nFIESTA, PathFinder demo study, codelist of subpopulations. https://gitlab.com/nfiesta/pathfinder_demo_study/-/ wikis/Codelists%20of%20subpopulations , 2023 (accessed 19 November 2024). 10] E. Tomppo , T. Gschwantner , M. Lawrence , R.E. McRoberts , National Forest Inventories, Pathways for Common Re- porting, Springer, Heidelberg Dordrecht London New York, 2010 . [11] C. Vidal , I. Alberdi , L. Hernández , J. Redmond , National Forest Inventories, Assessment of Wood Availability and Use, Springer, Cham, 2016 . 12] Forest Carbon Monitoring Product Platform. https://portal.forestcarbonplatform.org/ , 2023 (accessed 19 November 2024). https://doi.org/10.1093/forestry/cpp017 https://doi.org/10.1016/j.foreco.2009.10.011 https://doi.org/10.1111/geb.12125 https://doi.org/10.3390/rs8070565 https://doi.org/10.5194/essd-13-3927-2021 https://pathfinder-heu.eu/ https://doi.org/10.5281/zenodo.13143235 https://gitlab.com/nfiesta/pathfinder_demo_study/-/wikis/Codelist%20of%20target%20variables https://gitlab.com/nfiesta/pathfinder_demo_study/-/wikis/Codelists%20of%20subpopulations http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0010 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0010 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0010 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0010 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0010 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0011 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0011 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0011 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0011 http://refhub.elsevier.com/S2352-3409(25)00345-2/sbref0011 https://portal.forestcarbonplatform.org/ J. Miettinen, J. Breidenbach and P. Adame et al. / Data in Brief 60 (2025) 111613 15 [13] J. Miettinen, S. Carlier, L. Häme, A. Mäkelä, F. Minunno, J. Penttilä, J. Pisl, J. Rasinmäki, Y. Rauste, L. Seitsonen, Demonstration of large area forest volume and primary production estimation approach based on Sentinel-2 im- agery and process based ecosystem modelling, Int. J. Remote Sens. 42 (2021) 9467–9489, doi: 10.1080/01431161. 2021.1998715 . [14] Copernicus Land Monitoring Service, User manual, high resolution land cover characteristics, tree-cover/forest and change 2015-2018. https://land.copernicus.eu/en/technical-library/hrl-forest-2018 , 2021 (accessed 19 November 2024). [15] M.C. Hansen, P.V. Potapov, R. Moore, M. Hancher, S.A. Turubanova, A. Tyukavina, D. Thau, S. Stehman, S.J. Goetz, T.R. Loveland, High-resolution global maps of 21st-century forest cover change, Science (1979) 342 (2013) 850–853, doi: 10.1126/science.12446 . [16] nFIESTA, PathFinder demo study, scripts for auxiliary data extraction at plot positions. https://gitlab.com/nfiesta/ pathfinder_demo_study/-/blob/main/auxscripts/download_RS_data.R?ref_type=heads , 2023 (accessed 19 November 2024). [17] Forestry TEP Processing Platform. https://f-tep.com/ , 2024 (accessed 19 November 2024). https://doi.org/10.1080/01431161.2021.1998715 https://land.copernicus.eu/en/technical-library/hrl-forest-2018 https://doi.org/10.1126/science.12446 https://gitlab.com/nfiesta/pathfinder_demo_study/-/blob/main/auxscripts/download_RS_data.R?ref_type=heads https://f-tep.com/ Pan-European forest maps produced with a combination of earth observation data and national forest inventory plots 1 Value of the Data 2 Background 3 Data Description 4 Experimental Design, Materials and Methods 4.1 NFI Field Data Processing 4.2 Remote Sensing and Auxiliary Data Processing 4.3 Application of the kNN Approach Limitations Ethics Statement Data Availability CRediT Author Statement Acknowledgements Declaration of Competing Interest References