The PROFOUND Database for evaluating vegetation models and simulating climate impacts on European forests

Process-based vegetation models are widely used to predict local and global ecosystem dynamics and climate change impacts. Due to their complexity, they require careful parameterization and evaluation to ensure that projections are accurate and reliable. The PROFOUND Database (PROFOUND DB) provides a wide range of empirical data on European forests to calibrate and evaluate vegetation models that simulate climate impacts at the forest stand scale. A particular advantage of this database is its wide coverage of multiple data sources at different hierarchical and temporal scales, together with environmental driving data as well as the latest climate scenarios. Specifically, the PROFOUND DB provides general site descriptions, soil, climate, CO₂, nitrogen deposition, tree and forest stand level, and remote sensing data for nine contrasting forest stands distributed across Europe. Moreover, for a subset of five sites, time series of carbon fluxes, atmospheric heat conduction and soil water are also available. The climate and nitrogen deposition data contain several datasets for the historic period and a wide range of future climate change scenarios following the Representative Concentration Pathways (RCP2.6, RCP4.5, RCP6.0, RCP8.5). We also provide pre-industrial climate simulations that allow for model runs aimed at disentangling the contribution of climate change to observed forest productivity changes. The PROFOUND DB is available freely as a “SQLite” relational database or “ASCII” flat file version (at https://doi.org/10.5880/PIK.2020.006/; Reyer et al., 2020). The data policies of the individual contributing datasets are provided in the metadata of each data file. The PROFOUND DB can also be accessed via the ProfoundData R package (https://CRAN.R-project.org/package=ProfoundData; Silveyra Gonzalez et al., 2020), which provides basic functions to explore, plot and extract the data for model set-up, calibration and evaluation.

1 Introduction

Process-based models are key tools for understanding systems and forecasting climate change impacts in ecology and Earth system science (Schellnhuber, 1999). Vegetation is a crucial component of the Earth system, and forests are particularly relevant through their influence on hydrological and biogeochemical cycles, biodiversity and ecosystem services. Process-based vegetation models are used as diagnostic tools to disentangle the influence of different environmental and human drivers on biogeochemical cycling as well as vegetation structure from local to plot-level (Eastaugh et al., 2011; Fontes et al., 2010; Pretzsch et al., 2015; Tiktak and van Grinsven, 1995) to global scales (Chang et al., 2017; Ito et al., 2017). At the same time these models are also the main tools to project climate change impacts on vegetation under changing environmental conditions, again from local (Reyer, 2015; Rötzer et al., 2013) to global levels (Zhu et al., 2016).

With increasing model complexity, the inclusion of more and more processes and models being increasingly used as tools for making quantitative projections for policy and management, there is a strong need to install some quality control on their performance. A basic requirement would be that models are actually able to match observed data. Moreover, while informal methods for calibration and model comparisons were often used in the past, the community has shifted in recent years towards more formal statistical methods for such tasks (Dietze et al., 2013; Hartig et al., 2012), which creates a need for systematic benchmarking data. For all these tasks, the availability of a wide range of data types crossing different spatial–temporal scales is generally viewed as beneficial (Grimm and Railsback, 2012).

Figure 1 Location of forest sites and main tree species. Background shows the European forest cover after Brus et al. (2012).

The process of formal calibration, comparison and evaluation of complex vegetation models is often hindered by the availability and the harmonization of suitable data. The data necessary to drive a vegetation model are often complex and need to be compiled from different data sources (e.g. Bagnara et al., 2019). In particular for model comparisons, besides data for the evaluation of individual models, common input and driving data for process-based vegetation models are needed to ensure fair comparisons between the participating models. Although model comparisons have a long tradition in vegetation modelling (Cramer et al., 1999, 2001; Bugmann et al., 1996; Morales et al., 2005), they have often been limited by overall data availability and comparability. Common databases that are ready to use for thorough model evaluation would allow the community to gain a better appreciation of model differences, explore structural uncertainties and provide a basis for more systematic ensemble projections of climate impacts.

Recently, several initiatives have started compiling model evaluation, input or driving data for a wide range of applications of process-based vegetation models (Huntzinger et al., 2013; Kelley et al., 2013; Warszawski et al., 2014; Sitch et al., 2015; Collier et al., 2018). Although these initiatives have leveraged important scientific progress, many of them have focussed on the global scale, mostly providing evaluation, input and driving data from global products. Such global products generally lack the breadth and depth of process-level detail required to rigorously assess model performance at smaller scales as for example they lack long-term and detailed measurements of forest stand structure. The database for the project “Towards robust projections of European forests under climate change” (hereafter PROFOUND DB) described here aims to bring together data from a wide range of data sources to evaluate vegetation models and simulate climate impacts at the forest stand scale. It has been designed to fulfil two objectives:

• to allow for a thorough evaluation of complex, process-based vegetation models using multiple data streams covering a range of processes at different temporal scales and

• to allow for climate impact assessments by providing the latest climate scenario data.

Table 1 Overview of the data available in the PROFOUND DB. The years indicate the first and the last years for which data are available except for one-off measurements. The superscript letters indicate the temporal resolution of the data. O: one-off measurement(s); M = 30 min measurements; D: daily measurements; C: 8 d or 16 d composite; A: annual measurements.

The PROFOUND DB only provides data for individual forest stands but contains a number of elements that are designed to foster comparison of both global/regional models and local models. The climate data, for example, are provided locally (or bias-corrected using local data) in the same way that stand-scale vegetation models would need them and also extracted from global gridded datasets that global vegetation models would use. The PROFOUND DB is also designed to allow for disentangling of uncertainties that affect quantitative model predictions in ecology (see Lindner et al., 2014, and Dietze, 2017, for an explanation of different uncertainty types), for example by facilitating standardized evaluations of structural or process uncertainties via model comparisons. Model input and driver uncertainty are addressed through a wide range of climate data from different sources, covering the full range of Representative Concentration Pathways (RCPs). Collalti et al. (2018, 2019), for example, have used the PROFOUND DB to study the effects of thinning on carbon use efficiency across a combination of all four RCPs and five global climate models. Finally, parametric uncertainty can be assessed through the wide range of data that can be used for inverse calibration. In the following we describe the main components of the PROFOUND DB (Reyer et al., 2020) and an R Package (Silveyra Gonzalez et al., 2020) developed to explore the database and allow rapid and easy access for modellers.

2 The PROFOUND Database

2.1 Forest site selection and concept

The forest sites featured in the PROFOUND DB were selected to provide a wide array of data sources across a European gradient. We focussed in particular on providing long time series of tree- and stand-level growth and yield as well as carbon cycle data available from eddy-flux measurements because these variables are most commonly in calibrating and evaluating process-based vegetation models. The selected sites spread along a wide climatic gradient across Europe (Fig. 1, Table 3) and cover some of the most common European forest types, as well as the main central European forest management history of favouring monospecific, even-aged forests or mixtures of two tree species.

We compiled the data from existing data sources and collected the definitions of variables, their units and information about the main measurement methods from the site principal investigators (PIs) and from official descriptions of the data to harmonize the variables as much as possible. The overall guiding principle for the compilation of the data was to provide data that can be easily used by modellers for setting up and evaluating their models. In order to allow for data uncertainty to be reflected in model calibration studies, we also included uncertainty estimates for the measured data, such as those available for carbon flux measurements (see Sect. 3.2.9), wherever possible.

Table 2 Overview of the main site characteristics provided for each forest site in the PROFOUND DB.

^a According to ISSS-ISRIC-FAO (1998). ^b Depending on base saturation under or over 50 % with a 10–40 cm deep organic layer (see Pilegaard et al., 2003).

2.2 Data sources

The PROFOUND DB provides information on the site, soil and forest stand as well as data for climate, atmospheric CO₂ concentration, nitrogen deposition, carbon fluxes, atmospheric heat conduction and remote sensing at a range of different temporal resolutions (i.e. from 30 min to decadal measurements). Table 1 provides an overview of the different data types and their temporal resolution available in the PROFOUND DB. All variables available are listed in Tables S1–S13 in the Supplement. In the following we describe how the individual sub-datasets of the PROFOUND DB have been brought together and describe the key variables and characteristics of each dataset.

2.2.1 Site information

For each forest site, the PROFOUND DB contains information on general site characteristics such as coordinates, elevation and forest type (Table 2). There is also information on the potential natural vegetation and main tree species belonging to the regional flora (not shown).

2.2.2 Soil data

The description of the soil profiles contains information about physical and chemical properties of each soil horizon including the organic layer. Unfortunately the soil data are very heterogeneous for the sites, and considerable amounts of data are missing. In order not to lose the data that are available for only a subset of sites, we did not harmonize the individual variables, but for each site we provide the soil data in a consistent format. Despite these limitations, for most sites important soil data such as the depth of horizons, soil texture, bulk density, field capacity, wilting point, carbon and nitrogen content, and pH of the soil solution are available (see Table S2).

Table 3 Averages of the daily maximum temperature (T_max), daily minimum temperature (T_min), daily mean temperature (T_mean), annual precipitation sum (P), daily mean relative humidity (RH), daily mean air pressure (AP), annual sum of global radiation (R, direct + diffuse shortwave radiation) and daily mean wind speed (W) for each of the sites in the PROFOUND DB from five different sources: a locally observed climate and four different global reanalysis/observational products (GSWP3, Princeton, WATCH, WFDEI). The column “Year” indicates the years for which the mean climates have been calculated for the different sources. Please note that the two Solling sites have the same climate. NA: not available.

Figure 2 Change in mean annual temperature (T mean), annual precipitation sum (P) and annual sum of global radiation (R) over the time period 1950–2099 relative to the 1980–2005 average for the ISIMIPFT scenarios. Please note that the two Solling sites have the same climate.

2.2.3 Local climate

For every site we compiled the locally observed daily meteorological data, either from measurement towers or from nearby meteorological stations. These time series cover the main climatic variables required by vegetation models and different time periods for each site (Table 3). They represent the best possible climate information for each site and are most suitable for model simulations comparing simulation output to observations.

2.2.4 Reanalysis products

In order to cover longer historical time periods and to assess uncertainties due to the choice of different climate inputs, the PROFOUND DB also provides long historical daily climate time series for each of the sites extracted from four different global reanalysis/observational products:

• Princeton's Global Meteorological Forcing Dataset (PGMFD v.2, hereafter Princeton) from 1901 to 2012 by Sheffield et al. (2006);

• Global Soil Wetness Project Phase 3 (GSWP3) from 1901 to 2010 by Hjungjun Kim (personal communication, 2018, http://hydro.iis.u-tokyo.ac.jp/GSWP3/, last access: 5 June 2020);

• Water and Global Change programme (WATCH) from 1901 to 2001 by Weedon et al. (2011);

• WATCH-Forcing-Data-ERA-Interim (WFDEI) from 1901 to 2010 by Weedon et al. (2014).

Climate variables for the forest stands were extracted from the $\mathrm{0.5}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ grid cell of the global reanalysis/observational product in which the forest stand is located. The data are then kept at the original $\mathrm{0.5}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ resolution to allow for comparing the effects of choosing climate inputs for a vegetation model from a global reanalysis product as opposed to the local data presented in Sect. 3.2.3. The difference between the local data and the reanalysis data is most obvious for those sites located in complex, hilly terrain such as Collelongo or KROOF (Table 2). In these hilly locations the grid box average heights of the reanalysis products differ substantially from the heights of the site measurements.

2.2.5 Climate scenarios

The PROFOUND DB provides climate scenarios based on simulations performed for CMIP5 (https://www.wcrp-climate.org/wgcm-cmip/wgcm-cmip5; last accessed: 5 June 2020) that were bias-corrected and interpolated to a common grid resolution of $\mathrm{0.5}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ according to Hempel et al. (2013). The climate variables for each site available were extracted from the grid cell of the downscaled climate forcing dataset in which the forest plot is located. The data can be used in very different ways by the vegetation modelling community.

• The “ISIMIP Fast Track” scenarios (ISIMIPFT) consist of daily climate data available from five different global climate models (GCMs) (HadGEM2-ES, IPSL-CM5A-LR, MIROC-ESM-CHEM, GFDL-ESM2M and NorESM1-M.) for all four RCPs (Warszawski et al., 2014). The historical period lasts from 1950 to 2005 and then splits up into the four RCPs from 2006 to 2099 for each model. The RCPs cover future warming ranges of about 0–9 ^∘C in the late 21st century compared to the 1980–2005 average (Fig. 2). These ISIMIPFT data are best suited for scenario studies that require a large ensemble of GCMs and RCPs.

• The “ISIMIP2b” scenarios (ISIMIP2b) consist of daily climate data available from four different GCMs (IPSL-CM5A-LR, GFDL-ESM2M, MIROC5, HadGEM2-ES) for RCP2.6 and RCP6.0 (Frieler et al., 2017; Lange, 2018) as well as RCP4.5 and RCP8.5. The historical period lasts from 1861 to 2005 and then splits up into the four RCPs for each GCM from 2006 to 2099. The RCPs cover future warming ranges of about 1–9 ^∘C in the late 21st century compared to the 1980–2005 average (Fig. S1). For RCP2.6, RCP4.5 and RCP8.5 from IPSL-CM5A-LR, HadGEM2-ES and MIROC5, additional data are also available for the period 2100–2299. These long-term climatic pathways stabilize at around 1–2 ^∘C in the end of the 23rd century compared to 1980–2005 for RCP2.6, around 3–5 ^∘C for RCP4.5 and up to 16 ^∘C for RCP8.5. For all four GCMs, there are also time series of pre-industrial climatic conditions available from 1661 to 2299 (or 1661–2099 for GFDL-ESM2M), the so-called pre-industrial control run. The pre-industrial climates from each GCM for the time period 1661–1860 can be combined with the historical climates from 1861 to 2005 and any future time periods from the corresponding GCM to create a long-term time series of climate data from 1661 to 2299 (or 2099 depending on the GCM–RCP combinations) without almost any resampling (Frieler et al., 2017). The ISIMIP2b data are best suited to test the implications of long-term stabilization pathways and different degrees of warming relative to pre-industrial conditions in vegetation models.

• The “ISIMIP2b locally bias-corrected” scenarios (ISIMIP2bLBC) have the same structure as the ISIMIP2b data but have been bias-corrected using an improvement of the method of Hempel et al. (2013) as described in Frieler et al. (2017) and Lange (2017) and the local observed climatologies presented in Sect. 3.2.3. The ISIMIP2bLBC data are hence best suited for scenario studies that require climatic data to be as consistent as possible with the observational data (Fig. 3).

Figure 3 Change in mean annual temperature (T mean), annual precipitation sum (P) and annual sum of global radiation (R) over the time period 1661–2299 relative to the 1980–2005 average for the ISIMIP2b locally bias-corrected (ISIMIP2bLBC) scenarios. Please note that the two Solling sites have the same climate.

Figure 4 Global atmospheric CO₂ concentrations provided for all sites in the PROFOUND DB. The historical time period extends from 1765 to 2015 and the scenarios from 2005 to 2500 for each RCP.

2.2.6 Atmospheric CO₂ concentrations

Time series of atmospheric CO₂ concentrations are provided as annual, global data, hence as one time series for all sites of the PROFOUND DB assuming a well-mixed atmosphere. The historical time series of atmospheric CO₂ are based on global atmospheric CO₂ concentrations from Meinshausen et al. (2011) from 1765 to 2005 and have been extended for the period 2006–2015 with data from Dlugokencky and Tan (2014). The future annual atmospheric CO₂ concentrations follow the four different Representative Concentration Pathways (RCPs, RCP2.6, RCP4.5, RCP6.0 and RCP8.5) from 2016 to 2500 from Meinshausen et al. (2011). Figure 4 shows the historical increase in CO₂ concentrations since 1765 and the projected future emissions according to the different RCPs. From RCP2.6 till RCP8.5 the total level of CO₂ increases strongly, and also the date of stabilizing emissions is reached much later in RCP8.5. RCP2.6 is the only RCP that projects declining CO₂ levels in the long run.

2.2.7 Nitrogen deposition

The nitrogen deposition data, reported as total deposition of reduced and oxidized wet and dry nitrogen deposition, respectively, have been extracted for each site of the PROFOUND DB from two different datasets which serve different purposes.

• EMEP data. For detailed model evaluation studies that require the best possible estimates of local nitrogen deposition, we extracted data from the “Co-operative programme for monitoring and evaluation of long-range transmission of air pollutants in Europe” (EMEP) for the time period 1980–2014 (EMEP/CEIP, 2014a, b). Sea-salt-corrected data are available from 1980 to 1995 in 5-year steps and from 1986 to 2014 at annual time step and are derived by atmospheric transport modelling (Simpson et al., 2012).

• ISIMIP data. For model simulation studies, we also provide nitrogen deposition estimates based on atmospheric chemistry modelling for a historical time period (1861–2005) and four future scenarios, where nitrogen deposition follows the four RCPs. The data are further described in Lamarque et al. (2013a, b), sea-salt-corrected and consistent with the global nitrogen deposition data provided within ISIMIP (Frieler et al., 2017). The data are taken from the global dataset without further corrections and hence are not intended to represent realistic, local forecasts but rather rough estimates of future nitrogen projections.

For the 1980–2014 time period, the ISIMIP data are typically lower and less dynamic than the EMEP estimates (Fig. 5). However, while they do not seem suitable for historical model evaluations, they cover a much longer time period and are clearly interesting for scenario studies because they feature different nitrogen deposition pathways consistent with RCP climates and CO₂ pathways. It is also important to note that measured throughfall of NO₃ and NH₄ is on average lower than modelled total deposition, due to canopy uptake (Marchetto et al., 2020). Moreover, for the two Solling sites the data presented here are identical while in reality total N deposition rates in the spruce stand should be higher because of higher dry depositions. Actually, the ratio between Solling spruce and Solling beech is 1.4 for NH₄ throughfall fluxes, 1.6 for NO₃ throughfall fluxes, 1.4 for NH₄ total deposition and 1.4 for NO₃ total deposition, both using a canopy budget model (Ulrich, 1994) for the period 1980–2014. However, these ratios are not constant and show an increasing trend over time.

Figure 5 Total deposition of reduced (NH_x) and oxidized (NO_y) nitrogen (N) at each of the sites of the PROFOUND DB. The historical period for the EMEP data extends from 1980 to 2014 and for the historical ISIMIP data from 1861–2005. The future scenarios are available from 2006 to 2100 and follow the RCP2.6 and RCP6.0 scenarios. Please note that the two Solling sites have the same N depositions (see text for further explanations).

2.2.8 Forest inventory data

For each site, the PROFOUND DB provides information about the forest stand at tree and stand level (Table 4). The data are available for different time periods and have different measurement intervals, but generally they cover mostly the second half of the 20th century and the first decade of the 21st century (Table 1). The data also cover a wide array of height–age and diameter at breast height (DBH)–age relationships (Figs. 6–7). For seven out of nine sites individual tree DBH and height measurements are available. The time series length ranges between 15 and 65 years within the time period 1948–2015. For the Sorø site, the DBH and heights have been reconstructed from tree-ring data (Babst et al., 2014), and the full stand reconstruction is available from 1996 to 2010 at annual resolution (see Sect. S1 in the Supplement). Individual tree data allow analysis and comparison of model simulations with data on single-tree growth. From the tree data, we calculated a range of widely used stand variables (see Table S8). Additional stand-level data are available for some of the sites, such as leaf litter production or leaf area index, and have been included (see Table S8).

Table 4 Summary of the main stand variables for the forest stands in the PROFOUND DB. The first number in each cell indicates the value at the first measurement and the second number at the last measurement. The basal area-weighted mean height and DBH are shown for all stands with the exception of Le Bray for which the arithmetic mean height and DBH are shown (marked by asterisks). The numbers in brackets indicate different data availability for height than for the other variables.

Figure 6 Time series of tree diameter at breast height (DBH) versus age of the forest stands in the PROFOUND DB. The basal area-weighted mean DBH is shown for all stands with the exception of Le Bray for which the arithmetic mean DBH is shown (marked by asterisks). For Sorø, the DBHs have been reconstructed (see text in Sects. 4.9 and S1).

Figure 7 Time series of tree height versus age of the forest stands in the PROFOUND DB. The basal area-weighted mean height is shown for all stands with the exception of Le Bray for which the arithmetic mean height is shown (marked by ^*). For Sorø, the heights have been reconstructed (see text in Sects. 4.9 and S1).

2.2.9 Flux data

The carbon fluxes, i.e. net ecosystem exchange (NEE), ecosystem respiration (RECO) and gross primary production (GPP) are taken from the Tier One FLUXNET2015 dataset (http://fluxnet.fluxdata.org/, last access: 5 June 2020). We provide estimates of fluxes calculated using different estimates for gap-filled and partitioned fluxes to give a rough estimate of the uncertainty added to the long-term budgets in the process. NEE data are filtered using two different methods to calculate uStar thresholds (Barr et al., 2013, and a modified version of Papale et al., 2006; see also FLUXNET2015, 2017). Daytime (i.e. Lasslop et al., 2012) and night-time (i.e. Reichstein et al., 2005) refer to whether ecosystem respiration parameters were estimated from only night-time fluxes or also using daytime data (zero intercept of GPP light response curve). In many cases the number of accepted night-time fluxes is low and the temperature range is narrow, which leads to high uncertainty in the estimated respiration. This can be improved by also using daytime fluxes. On the other hand in the daytime method the uncertainties of photosynthetic light, temperature and possible vapour pressure deficit (VPD) responses may be attributed to respiration parameters. Further information about the daytime and night-time methods is available in Lasslop et al. (2010) and Reichstein et al. (2005) and also FLUXNET2015 (2017). We also extracted different uncertainty estimates for each variable. Additionally, we provide time series of the sensible and latent heat flux, soil (soil water and soil temperature) and meteorological variables at a 30 min time resolution from the FLUXNET2015 database including measurement uncertainty estimates. Table 5 provide an overview of the main carbon fluxes at each of the sites featured in the PROFOUND DB. Tables S9 and S11–S13 provides the full list of available variables.

Table 5 Summary of the observed carbon fluxes at the sites in the PROFOUND DB. Shown is the range (min and max) and the average (in brackets) of the annual sums in the observational period. All data are estimates based on the CUTRef method with daytime data included for RECO and GPP. GPP is expressed with negative values because it is considered a downward flux from the atmosphere. Likewise, negative NEE values indicate a carbon sink and positive values a carbon source.

^* The year 2007 is without data for Hyytiälä and the year 2002 for Le Bray.

Table 6 Summary of the remote sensing data included in the PROFOUND DB. VIS, NIR and SWIR are the visible, near-infrared and shortwave infrared regions of the electromagnetic spectrum. NDVI: normalized difference vegetation index; EVI: enhanced vegetation index; FPAR: fraction of photosynthetically absorbed radiation; LAI: leaf area index; GPP: gross primary productivity; NDWI: normalized difference water index; AR: angle at red; ANIR: angle at NIR; AS1: angle at shortwave infrared 1; AS2: angle at shortwave infrared 2; SANI: shortwave angle slope index; SASI: shortwave angle slope index.

2.2.10 Remote sensing data

The PROFOUND DB includes remote sensing information at different spatial scales and temporal frequencies, specific for each product. We included five MODIS products (ORNL DAAC, 2008a–e) and several vegetation indices calculated from the surface reflectance data for each of the forest sites. The original MODIS scenes are available at the NASA Land Processes Distributed Archive Center (LP DAAC) (https://lpdaac.usgs.gov/, last access: 5 June 2020). The specific time series included in the PROFOUND DB were downloaded from the Land Product Subset Web Service of the Oak Ridge National Laboratory Distributed Active Archive Center (ORNL DAAC) (https://daac.ornl.gov/MODIS/, last access: 5 June 2020). The ORNL DAAC MODIS subsetting web service is implemented to allow users access to massive amounts of remote sensing data (Santhana-Vannan et al., 2011). In addition, a second set of vegetation indexes was calculated from the reflectance values. A summary of this information is shown in Table 6. The full list of variables and how they were aggregated is provided in Table S10.

The main difference among the forest sites is the data quality, which is highly dependent on the presence of clouds. When possible, low-quality observations have been substituted by interpolated values, otherwise the cell was left blank. In any case the alteration of the original data was minimal. It is also important to note that the size of the pixel is large compared to the plot size of the forest stands, which means the pixel data also contain other vegetation types than the ones present at the sites.

Three general types of data are included: (1) geophysical variables as measured from the MODIS sensor, i.e. reflectance and temperature; (2) spectral indexes derived directly from reflectance values at different wavelengths; and (3) vegetation properties (i.e. FPAR, LAI, GPP and net photosynthesis) as estimated from physical variables through a range of models.

Although the MODIS sensor acquires daily information, the PROFOUND DB includes only composite data; that is, for each pixel the best value during a period of time (8 or 16 d) is selected as being representative of that specific period. Spatial resolution is also specific for each product and is dependent on the physical and technical limitations in the acquisition process of the variables involved in the product computation.

Table 7 Generic future management scenarios for the main tree species featured in the PROFOUND DB.

The NDVI and EVI at 250 m spatial resolution coming from the MOD13Q1 product were calculated from the visible and near-infrared spectral regions. A temporal frequency (16 d composite) was chosen to minimize the effect of clouds. The EVI index was developed to correct for atmospheric and background effects so that it shows a larger dynamic range in areas with high vegetation density (Didan et al., 2015).

The spectral profiles in the whole optical domain (i.e. 459–2155 nm) for each 8 d composite are represented by the surface spectral reflectance at seven wavelengths coming from the MOD09A1 product at 500 m spatial resolution. The criteria for the compositing process are low cloudiness, cloud shadows and low solar zenith angle; when several of these criteria are fulfilled the selection is based on the minimum value in the blue band (Vermote et al., 2015).

The second set of spectral indexes was computed from the MOD09A1 product. The indices based on the spectral shape have the advantage of combining information on three bands instead of two, and when the bands used are located in the SWIR region relevant information related to water is captured (Palacios-Orueta et al., 2005; Khanna et al., 2007; Palacios-Orueta et al., 2012).

Table 8 Planting information for the sites included in the PROFOUND DB. The numbers in brackets indicate plausible ranges (NA: not available).

LAI is defined as the one-sided green leaf area per unit ground area in broadleaf canopies and as one-half the total needle surface area per unit ground area in coniferous canopies. The FPAR is the fraction of photosynthetically active radiation (400–700 nm) that is absorbed by the canopy (Myneni, 2015). Gross primary productivity and net photosynthesis estimations are based on the light use efficiency (LUE) concept (Monteith, 1972) using satellite-derived FPAR (from MOD15) and independent estimates of PAR, besides other types of ancillary data. These are highly aggregated variables that have gone through several modelling steps already. Detailed information on the model and information sources used can be found in Running and Zhao (2015).

3 Description of the forest sites

The most northern site is Hyytiälä in Finland with a boreal climate, while the most southern sites are Le Bray in France and Collelongo in Italy with an oceanic and Mediterranean montane climate, respectively. All other sites represent temperate climatic conditions ranging, however, from oceanic (Belgium, Denmark) to temperate (France, Germany) to sub-continental (Czech Republic). Unfortunately, sites representing more continental and (east) Mediterranean forests from southern and southeastern Europe are missing.

3.1 Bily Kriz (CZ)

The Bily Kriz site belongs to the ICP Forests Level II network and is a FLUXNET site located in the Moravian–Silesian Beskydy Mountains, Czech Republic, at an altitude of 875 m a.s.l. The climate is temperate with an annual mean temperature of 7.4 ^∘C and an annual precipitation sum of 1434 mm over the 2000–2008 period. The soil is classified as a Haplic Podzol. The site is typical for mountain regions of temperate Europe such as the Black Forest, Bohemian Forest Šumava and forested Carpathians (Hercynian (spruce–)fir–beech forests) but also the higher mountain belts in the (sub-)Mediterranean. Stand-forming tree species for such sites are Fagus sylvatica, Abies alba and Picea abies. Currently, a large part of mixed mountain forests are strongly managed for timber production. The main tree species occurring in Bily Kriz are Picea abies, rarely with a small proportion of Fagus sylvatica. The stand data represent an (even-aged) Picea abies monoculture with a mean DBH of 19 cm (year 2015). The potential vegetation belongs to the geobiocoene-type groups: Abieti-fageta (5AB3) – Abies alba Mill. + Fagus sylvatica L. with understory: Calamagrostis arundinacea (L.) Roth, Oxalis acetosella L., Vaccinium myrtillus L., Deschampsia flexuosa (L.) Trin. More information about the site can be found in Kratochvilova et al. (1989) and Meteorological yearbook (2012).

3.2 Collelongo (IT)

The experimental site of Collelongo is located in Selva Piana, a pure Fagus sylvatica forest in Collelongo (AQ, central Italy) at 1560 m a.s.l. Located 100 km from Rome, it is one of the first Italian sites of the ICP network and also part of the ILTER international network. The climate is Mediterranean montane, with a mean annual temperature of 7.2 ^∘C and a mean annual precipitation of 1179 mm in the period 1996–2014. Bedrock consists of Cretaceous limestone. Soil depth exhibits high spatial variability ranging from 40 to 100 cm and is classified as a Humic Alisol (Chiti et al., 2010) or Dystric Luvisol according to the FAO classification. The stand is a typical Apennine beech forest dominated by Fagus sylvatica with sporadic trees of Taxus baccata. The phytosociological association is Polysticho – Fagetum (Feoli and Lagonegro 1982). Currently, Collelongo constitutes a managed Fagus sylvatica stand with mean DBH of 25 cm in 2012. In the area around the eddy-flux tower there are only Fagus sylvatica trees. Moreover the footprint of the tower is totally included in the Fagus sylvatica forest. More information about the site can be found in Chiti et al. (2010), Collalti et al. (2016) and D'Andrea et al. (2019).

3.3 Hyytiälä (FI)

The most northern site included in the PROFOUND DB is the ICP Forests Level II site Hyytiälä, Finland. It is also a FLUXNET site and the coldest site with an annual temperature of 4.4 ^∘C and 604 mm annual precipitation during the 1996–2014 period and lies at 185 m a.s.l. The soil is classified as a Haplic Podzol. Picea abies is the naturally dominant tree species building Fennoscandian moss-rich spruce forests with Pinus sylvestris. A Pinus sylvestris stand was sown in 1962, today with admixtures of Picea abies and hardwood species (Betula pendula, Betula pubescens and Populus tremula). Mean DBHs were 17 cm for P. sylvestris, 5 cm for P. abies and 7 cm for hardwood species in the year 2008. More information about the site can be found in Haataja and Vesala (1997), Rannik et al. (2004), Vesala et al. (2005), Ilvesniemi et al. (2009), Mammarella et al. (2009), and Ilvesniemi et al. (2010).

3.4 KROOF (DE)

The KROOF forest belongs to the “Kranzberg Forest Roof Experiment” of the Technical University of Munich (TUM) and the Helmholtz Zentrum München. The site is located close to Freising, Germany, in the Kranzberger Forst in 502 m a.s.l. (wc-alt.). Mean annual temperature is around 8.2 ^∘C, and annual rainfall is around 849 mm during the period 1998–2010. The soil type, Luvisol, is typical for the region. The potential natural vegetation is (sessile oak–) beech forest (Fagus sylvatica, Quercus petraea, Quercus robur). The establishment of the research plot dates back to 1992. The mixed stand comprises large groups of Fagus sylvatica surrounded by Picea abies with mean DBHs of 26 and 33 cm in 2010, respectively. Other occurring species are Acer platanoides (20 cm), Pinus sylvestris (31 cm), Larix decidua (26 cm) and Quercus robur (29 cm). More information about the site can be found in Pretzsch et al. (1998, 2014) and Matyssek et al. (2014).

3.5 Le Bray (FR)

The ICP Forests site Le Bray is located 20 km southwest of Bordeaux, France, at an altitude of 61 m a.s.l. Mean annual temperature is about 13.4 ^∘C, and precipitation is 920 mm during the 1996–2008 period, constituting a moderate oceanic climate. The soil type is Arenosol (sandy and hydromorphic podzol), which is one of the most common soils in the region. The natural vegetation is formed by deciduous broadleaf forests such as pedunculate oak forests (Quercus robur), partly with Quercus pyrenaica, Quercus suber and Pinus pinaster. The first measurements were made in 1986 in the monospecific planted Pinus pinaster stand. The site experienced a storm in 1999 and lost a large number of trees. In 2009, the mean DBH was 35 cm. The final clear cut of the site occurred at the beginning of 2009. More information about the site can be found in Porté and Loustau (1998), Bosc et al. (2003) and Berbigier et al. (2001).

3.6 Peitz (DE)

Peitz is a long-term research plot in eastern Brandenburg, Germany. The site lies at about 50 m a.s.l. The annual rainfall amounts to more than 608 mm, and annual mean temperature is around 9.2 ^∘C during the 1901–2010 period. The soil type is a Dystric Cambisol. The potential natural vegetation is a south Scandinavian, east central European dwarf shrub and lichen-rich pine forest mix (Pinus sylvestris), partly with Quercus robur in the understorey, with Vaccinium vitis-idaea, Calluna vulgaris, Cladina spp. and Dicranum polysetum on sandy soils and siliceous rocks. The forest is a pine forest (Pinus sylvestris) with a mean DBH of around 23 cm and a stand height of 17 m in 2011. The understorey consists partly of Quercus robur. Measurements were started in 1948. More information about this site can be found in Riek and Stähr (2004) and Noack (2011, 2012) and about the climate data in Gerstengarbe et al. (2015).

3.7 Solling beech (DE)

Solling 304 is a long-term intensive forest monitoring plot (Level II) of the ICP Forests network in central Germany. The plot is also part of the LTER (site LTER_EU_DE_009) and of the permanent soil monitoring programme of the state of Lower Saxony. The site is situated in the centre of the Solling plateau at an elevation of about 500 m a.s.l. The mean temperature was around 6.8 ^∘C and the mean annual rainfall amounted to 1113 mm during the period 1960–2013. The bedrock consist of Triassic sandstone covered with a 60 to 80 cm deep solifluction layer of loess material from which the soil, classified as an Haplic Cambisol, has developed. The humus type is a typical Moder. The tree layer consists only of European beech (Fagus sylvatica L.). Oxalis acetosella and Luzula luzuloides are the major species of the sparse ground vegetation. Actual vegetation was assigned to the Luzulo-Fagetum typicum and is close to the potential natural vegetation. The forest is a 168-year-old stand with a mean DBH of 50 cm and a mean height of 30.7 m in 2016. More information about the site can be found in Meiwes et al. (2009), Meesenburg et al. (2009, 2016), Panferov et al. (2009), Le Mellec et al. (2010), and Fleck et al. (2016).

3.8 Solling spruce (DE)

Solling 305 is also a long-term intensive forest monitoring plot of the ICP Forests Level II network in central Germany. As the Solling beech site it belongs to the LTER (site LTER_EU_DE_009) and is a permanent soil monitoring plot of the state of Lower Saxony. It is situated close to the Solling beech site at an elevation of about 508 m a.s.l. and has similar site conditions as the Solling beech stand. Potential natural vegetation is a Luzulo luzuloido Fagetum. Dominant species of the actual ground vegetation are Vaccinium myrtillus, Polytrichum formosum and Deschampsia flexuosa (Bolte et al., 2004). The forest is a 133-year-old Norway spruce (Picea abies) stand with a mean DBH of 46.6 cm and a mean height of 33.1 m in 2016. More information about the site can be found in Le Mellec et al. (2010), Bonten et al. (2011), Meesenburg et al. (2016), Fleck et al. (2016) and Wegehenkel et al. (2017).

3.9 Sorø (DK)

The ICOS site Sorø (DK-Sor in the FLUXNET and ICOS databases) is located in Denmark at an elevation of 40 m a.s.l. The climate is warm temperate and fully humid with a mean annual temperature of 9 ^∘C and annual precipitation sum of 774 mm during the period 1996–2010. The soil has been classified as Alfisols and Mollisols. Potential natural vegetation is deciduous broadleaved forest dominated by Fagus sylvatica. Other species occurring in the area are Fraxinus excelsior, Larix decidua, Picea abies, Quercus spp. and Acer spp. However, the region is mostly used as cropland. Data on tree DBH are reconstructed from tree-ring measurement (Babst et al., 2014) and historical management information for the time period from 1994 to 2017. Stand data are derived from these data for the time period from 1994 to 2017 (see Sect. S1). The mean DBH of this Fagus sylvatica stand was 41 cm in the year 2017. More information about the site can be found in Ladekarl (2001), Pilegaard et al. (2003, 2011) and Wu et al. (2013). More information about the site can be found in Ladekarl (2001), Pilegaard et al. (2003, 2011) and Wu et al. (2013).

4 Forest management of the sites

The sites available in the PROFOUND DB are managed forests, and the historic management can be derived from the tree and stand-level data (in terms of reduction of stem numbers). However, for future scenario studies, generic, simple management and planting guidelines are available (Tables 7–8). This future management corresponds best to “intensive even-aged forestry” as defined by Duncker et al. (2012).

5 The PROFOUND R package (ProfoundData)

The ProfoundData R package provides functions to access the PROFOUND DB (Figs. S2 and S3). The ProfoundData package plus a detailed vignette explaining the functionalities are available on CRAN (https://CRAN.R-project.org/package=ProfoundData, last access: 5 June 2020). The ProfoundData package serves as an interface for users that want to access the PROFOUND DB as a relational database via the R statistical software (R Core Team, 2016). The following main functions are included to achieve this goal:

• “getData” to download data (data can be downloaded for one forest site and one underlying dataset at a time);

• “browseData” to check the available forest sites, datasets, variables for a dataset, datasets for a forest site and the database version, metadata, data policy and original data source;

• “plotData” to quickly inspect any variable of the datasets visually;

• “summarizeData” to summarize data from the database;

• “queryDB” to pass self-defined queries;

• “writeSim2netCDF” to write netCDF files and can be used to convert data (and other files such as model simulation output) into netCDF files.

While the ProfoundData R package is meant to provide easy access to the PROFOUND DB, the database is also fully functional without the R package.

6 Data availability

The PROFOUND Database (https://doi.org/10.5880/PIK.2020.006/, Reyer et al., 2020) is available under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). The PROFOUND R Package (ProfoundData, https://CRAN.R-project.org/package=ProfoundData, last access: 5 June 2020, Silveyra Gonzalez et al., 2020) is available via a GLP3 license. An earlier version of the database, including an outdated reconstruction of the Sorø tree data has been published as Reyer et al. (2019).

7 Conclusions

A wide range of data are needed to properly evaluate complex process-based vegetation models. The PROFOUND database compiles data from soil, climate, stand and flux measurements with data from remote sensing, atmospheric nitrogen modelling and climate modelling. Moreover, by providing data at $\mathrm{0.5}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ grid level plus locally bias-corrected climate data, the datasets can be used to compare local forest models to global vegetation models. The PROFOUND database thus facilitates model evaluation, calibration, uncertainty analysis and model intercomparisons, highlighting the immense value of long-term environmental monitoring data for robust inferences about causal processes and future dynamics of forests.

Appendix A: List of FLUXNET sites

Table A1 List of FLUXNET sites used in PROFOUND DB.