Fine earth soil bulk density at 0.2 m depth from Land Use and Coverage Area Frame Survey (LUCAS) soil 2018

Soil Bulk Density (BD) is an extremely important variable because it is an important site characterization parameter, and it is highly relevant for policy development because it is mandatory for calculating soil nutrient stocks. BD can influence soil chemical properties, land‐use planning and agronomic management. The 2018 Land Use and Coverage Area Frame Survey (LUCAS) saw the unprecedented collection of BD core analysis in a subset of the locations in Europe and the United Kingdom where soil physical and chemical properties were analysed in the 2009 and the 2015 sampling campaigns. Here, we integrated the LUCAS 2018 BD sampling campaign with the mass fraction of coarse fragments previously determined in LUCAS 2009–2015 in order to provide a dataset of the volume fraction of coarse fragments and the BD of the fine earth and improve soil organic carbon (SOC) stock estimation accuracy for topsoil. BD data sampled at 0–10 and 10–20 cm were averaged to harmonize the BD with the mass fraction of coarse fragments measured in 2009, 2012 and 2015. Samples were from cropland, grassland and woodland soils, which accounted for 41%, 21% and 30%, respectively, of the total number of selected sites (n = 6059); ‘bareland’, and ‘shrubland’ accounted for 3% of the sites each, whereas ‘artificial land’ accounted for <1%. Only six samples were classified as ‘wetland’. The dataset was produced assuming the mass density of the coarse fraction to be constant across all LUCAS soil samples. We also estimated the SOC stocks associated with LUCAS 2018 BD and SOC content measurements and showed that correcting the BD by the coarse mass fraction instead of the coarse volume fraction generates SOC stock underestimation. We found the highest deviations in woodlands and shrublands. We showed that, when SOC stock is computed with coarse mass fraction, the error compared with the computation by volume may vary depending on the SOC and coarse mass fraction. This may imply a SOC stock underestimation for European soils. This dataset fits into the big framework of LUCAS soil properties monitoring and contributes both to soil awareness and soil research and assessments, which are two important objectives of the Soil Strategy and the European Soil Observatory (EUSO).


| INTRODUCTION
The bulk density (BD) is a physical soil property highly dependent on soil particle size and soil organic carbon (SOC) and strongly influenced by the soil management (Ballabio et al., 2016;Fern andez-Ugalde & T oth, 2017;Rabot et al., 2018). BD is linked to soil functionality, comprising the rooting capacity and macro-and microporosity (Gao et al., 2017;Lipiec et al., 2006;Rawls et al., 1998;Zeri et al., 2018) that allow the circulation of soil solution and soil aeration (Håkansson & Lipiec, 2000). Soil BD is calculated as the dry weight of soil divided by its volume.
Soil BD is normally expressed as g cm À3 or Mg m À3 (SI). Along with SOC concentrations, soil BD is necessary to calculate SOC stocks (Minasny et al., 2013) and to assess soil compaction over time (Bogunovic et al., 2020;Katuwal et al., 2020).
Routinely measured properties in physical soil analysis include textural information and chemical properties such as nitrogen (N) content and organic carbon (SOC) content. To convert the soil nutrients' concentrations into stocks, it is necessary to convert the soil volume into a weight, and for this task, BD of the fine earth fraction (particles <2 mm) is required (Chen et al., 2018;Nasta et al., 2020;Van Looy et al., 2017;Veronesi et al., 2014). Poeplau et al. (2017) highlighted the importance of (i) correcting the BD measures obtained from soil cores to account for the mass and volume of coarse fragments (rock fragments that do not contain organic carbon and do not belong to the fine earth fraction) and (ii) considering the volume fraction occupied by the coarse fragments when calculating SOC stocks from BD and SOC content measures (Poeplau et al., 2017). The ISO standard used for the LUCAS 2018 BD laboratory analysis, ISO 11272:2017, considers the whole soil. However, this BD measure has to be corrected to take into account the presence of rock fragments. This requires knowing the density of coarse fragments, which is usually unknown but can be approximated.
Legacy data often allow for non-paired sample comparisons, which are performed in local and regional contexts for both BD and SOC and other soil properties (Schillaci et al., 2019;Siles et al., 2022). Presently, research in paired sites is promoted in the national and LUCAS soil modules, which is ongoing; we hope there will be a direct comparison in the near future. The European Soil Observatory (EUSO) and the European Soil Data Centre (ESDAC) are making their best efforts to democratize the EU soil data and knowledge and make it available to all stakeholders (Montanarella & Panagos, 2021;Panagos, van Liedekerke, et al., 2022). The aim of this work is to provide a ready-to-use dataset of fine earth's BD values for the Land Use and Coverage Area Frame Survey (hereafter LUCAS) 2018 Topsoils by modelling the effect of the coarse fragments volume derived from LUCAS coarse fraction data taken in 2009 and 2015. As an approximation, we assume the coarse fragments density to be constant across all samples. These data can be used to approximate SOC stocks from SOC contents measured in the previous LUCAS soil surveys. Soil BD data were gathered from the LUCAS Topsoil 2018 soil module. Coarse mass fractions were taken for the same sites from LUCAS Topsoil 2009 and LUCAS Topsoil 2015 (using the unique soil identifier). Coarse volume fraction data are not directly available and were estimated.
Finally, we simulated the potential deviation in SOC stock estimations generated by correcting the BD by the mass fraction instead of the volume fraction of coarse fragments. Since both the SOC and the BD are influenced by the land cover, we verified whether the deviations in SOC stock estimations are land-cover dependent.

Highlights
• BD data sampled at 0-10 and 10-20 cm were averaged to join the measurement with past LUCAS data. The LUCAS was designed to collect statistical information on land use and land cover over the territory of the EU from 2006. From the year 2009, an additional component (LUCAS Soil) was added consisting of stratified soil sampling for monitoring purposes. Soil samples and supporting data were collected by direct observations of about 22,000 points (a similar number was also collected by the 2015 LUCAS survey) by surveyors on the ground. The initial objective for the LUCAS soil survey was to collect data on soil physical and chemical properties, such as texture and organic carbon (SOC), with an emphasis on agricultural soils. Over time, the scope of the LUCAS soil survey was broadened, and additional parameters were collected and analysed. For the chemical and physical laboratory analysis, composite samples of approximately 500 g are taken from five subsamples collected with a spade at each LUCAS point. The first subsample is used to report the location coordinates, the other four subsamples were collected at a distance of 2 m, following the cardinal directions (North, East, South and West). In the exact place of sampling, stones (>6 cm) (FAO, 2006), plant residues, grass and litter were removed from soil surface by raking with the spade. The five subsamples in the bucket were mixed with a trowel. Aliquots (about 500 g) of the mixed soil are taken with a trowel from the bucket, placed in a plastic bag, and labelled to derive the composite sample. Soil samples were allowed to air dry before the bags were sealed. Based on the 2009 data, topsoil texture has been sampled in about 20 thousand locations and subsequently mapped for the EU (EU 26) with a nominal pixel resolution of 500 Â 500 m (Ballabio et al., 2019). Furthermore, SOC data are available from both the 2009 and the 2015 sampling campaigns and are spatially available at the same spatial resolution as the fine earth fraction (Ballabio et al., 2016). In LUCAS 2018, soil sampling was carried out in all EU Member States using the same set of 25,947 locations that were targeted in 2015 (Fernandez-Ugalde et al., 2022). In 65% of these locations, samples were to be taken following the standardized sampling procedure of previous surveys (Orgiazzi et al., 2015), in which a spade was used to collect a sample from a depth of 20 cm (Fern andez-Ugalde et al., 2020). In the year 2018, the monitoring scheme included the determination of the BD from a depth of 0-10 and 10-20 cm 3 in around 35% of the locations (approximately 9000 points) using metallic rings to collect soil cores. The total number of samples collected in the survey was 18,984, of which 18,744 locations were sampled at 0-20 cm depth. In the remaining 381 locations, due to particular field conditions (land ownership, meteorological conditions during the survey, and difficulties in reaching the locations), surveyors collected samples only from a single depth (i.e., 0-10, 10-20, or 20-30 cm). As for the LUCAS 2009, 2012 and 2015 surveys, all samples were analysed for physical and chemical properties in a single laboratory using the same analytical methods. During the 2018 survey in 6269 locations, BD samples were collected using metal cores. A stratified random sampling (Brungard et al., 2015;Minasny & McBratney, 2006) approach was used to select candidate points. This sampling design reflects diversity in soil texture and organic carbon content, land use and land cover, topography and soil type. The collection of BD samples was carried out in the same locations as the soil biodiversity samples to search for possible relationships between these properties. From the original number of samples foreseen, only 71% were sampled compared with 85% and 95% in LUCAS 2015 and LUCAS 2009/12 surveys, respectively. Unfortunately, due to logistic changes, extreme weather and soil conditions, the presence of potentially dangerous animals, and limited access due to the presence of fences in fields, access to the designated sites was difficult. The percentage of points sampled was below 50% in Germany, Croatia, Ireland, Malta, the Netherlands, Romania and the United Kingdom. The low rates of sampling were mainly attributed to legal issues related to neglected access in private in Germany, while in the UK, denial of access by landowners reflecting the Brexit referendum was an issue in many cases.

| Sample collection
For determining BD, soil cores were collected from 0-10 and 10-20 cm depths. With the common sampling procedure, the first sample is taken at the geo-referenced point location. Before taking the soil cores, stones (>6 cm) (FAO, 2006), vegetation residues, grass and litter were removed from the soil surface by raking with the spade as in the common sampling procedure. After the cleaning of the soil surface, five soil cores were taken from 0 to 10 cm depth with a metallic ring of 100 cm 3 at each LUCAS point. The other four BD soil cores were taken at a distance of 2 m following the cardinal directions (North, East, South and West). The procedure consisted of gently driving the metallic ring into the soil profile generated by the spade hole and pushing the ring using a wooden block with a mallet. This avoided the compaction of soil. The ring was removed from soil with the help of a spade placed underneath the ring. The excess soil around the ring was removed with a knife, and the soil core was pushed into a labelled plastic bag. The procedure was repeated in the four cardinal directions, and the soil cores collected were placed in the same labelled plastic bag. In the end, five soil cores of known volume were taken at 0-10 cm depth. After sampling of the 0-10 cm depth was completed, and placed into a plastic bag, the 10-20 cm depth was carried out following the same procedure; for additional details please consult fig. 3 of the technical report (Fernandez-Ugalde et al., 2022). The soil cores were then allowed to air-dry, and their weight was again recorded. The plastic bags were then sealed for their transportation to the laboratory. In Portugal, due to time restrictions, a minimum of three cores were collected from each depth.

| Core analysis
Samples were analysed in a single laboratory (SGS Hungary) for each property. BD was determined with soil cores, and it was determined in the 2018 survey for the first time. Before the analyses, a subsample of the soil cores taken at each depth was oven-dried and the weight recorded to determine BD from 0-10, 10-20, and 20-30 cm depths (the latter only for samples from Portugal). The soil cores from 0-10 and 10-20 cm depths were then mixed to derive a composite sample from 0 to 20 cm depth for its analysis. The soil cores collected from 20 to 30 cm depth in Portugal were kept apart and analysed only for organic carbon content. The ISO standard used to analyse soil properties, and in particular the BD, was the adapted ISO 11272:2017 (https://www.iso.org/standard/68255.html), calculated from the mass and the volume of sole cores taken with rings of known volume.

| Data selection
BD data sampled at 0-10 and 10-20 cm were averaged to harmonize the BD with the mass fraction of coarse fragments measured in 2009, 2012 and 2015. BD data collected at 20-30 cm depth (n = 142) were excluded from the analysis because all other LUCAS chemical and physical properties are not available for this depth interval yet. When one of the two BD measurements was missing (0-10 or 0-20 cm), we used the available BD measurement, assuming a uniform BD along the first 20 cm of topsoil. This concerned 26 samples for which the 0-10 cm BD measure was not available and 486 samples for which the 10-20 cm BD measure was not available.
The mass fraction of coarse fragments was sampled in LUCAS samples for each location at the time of the first data collection (2009 or 2015). According to the LUCAS specification for the field sampling procedures (Fernandez-Ugalde et al., 2022;Jones et al., 2020;Toth et al., 2013), samples were taken within a radius of 12.5 m from the original coordinates. This happened in the majority of cases, when there were no fences or other limitations on accessing the site. A database 'join' was performed to link the coarse fragments data and the BD using the unique identifier soil ID. We obtained a total of 6146 data points with BD and coarse fragments fraction data.
To avoid low-confidence data, we removed 47 samples that had BD <0.1 g cm À3 and SOC content <170 g C kg À1 (Sinclair et al., 2020), 39 samples that had BD v > 2 g cm À3 and a sample that was erroneously recorded to have negative BD. We obtained a total of 6059 samples.
For the data analysis only, we discarded one site whose land cover was classified as 'water'.

| Calculation of the bulk density of fine earth associated with LUCAS 2018 soil bulk density measures
We define BD sample as the BD measured with the soil core expressed in g.cm À3 , equal to the ratio m tot =V tot , where the total mass m tot ¼ m fine þ m coarse is the sum of the mass of fine earth (m fine ) and the mass of coarse fragments (m coarse and V tot ) is the fixed core volume, occupied by fine earth, coarse fragments and pores.
As an approximation, we attribute the pore volume completely to the fine earth fraction. Thus, we define the apparent volume of fine-earth fraction (V fine ) as the real volume of the fine earth plus the volume of the pores, and we define V coarse the real volume of coarse fragments, so that We define BD fine ¼ m fine =V fine the BD of the fine earth only.
BD fine is to be used for the calculation of SOC stocks from SOC concentration measures (grams of C per kilogram of fine earth), as the coarse rock fragments are not expected to contain organic carbon (Poeplau et al., 2017).
The values of m fine and V fine are not available in the LUCAS 2018 dataset; however, they can be calculated from the values of V tot , coarse massfraction , BD sample and ρ coarse , with ρ coarse the actual density of coarse fragments expressed in g.cm À3 . V tot is the fixed volume of the core; coarse massfraction is recovered from LUCAS 2009 and 2015; and BD sample is in LUCAS 2018.
The coarse fragments density ρ coarse is unknown in LUCAS but can be approximated to 2.6 g/cm 3 (Don et al., 2007).
Indeed, the mass of fine earth is equal to: and the mass of coarse fragments is equal to: but also: so that we can write: and by difference: Thus, BD fine is equal to: We calculated the approximated BD fine values associated with the LUCAS 2018 BD sample measures from Equation 6 using the coarse massfraction values from LUCAS 2009 and LUCAS 2015, and the approximation ρ coarse = 2.6 g/cm 3 .

| Conversion from mass fraction to volume fraction of coarse fragments using LUCAS 2009/2015 and LUCAS 2018 bulk density measures
To calculate SOC stocks, one needs to correct for the fraction of volume fraction occupied by coarse fragments coarse volumefraction (Poeplau et al., 2017). This measure is not available in LUCAS, but it can be recovered from the measure of coarse massfraction and from the approximated values of BD fine and ρ coarse .
Indeed, we have that.
with V coarse ¼ m coarse =ρ coarse and V fine ¼ m fine =BD fine and where the total pore volume is attributed to the V fine . So Equation 7 can be rewritten as: And we have that m coarse ¼ m coarse þ m fine ð Þ Â coarse massfraction and m fine ¼ m coarse þ m fine ð Þ Â 1 À coarsejmassfraction ð Þ , so we can write This can be rewritten as We calculated the approximated coarse volumefraction values associated with the LUCAS 2009/2015 measures of coarse massfraction from Equation 11 using the calculated BD fine approximated values and ρ coarse = 2.6 g/cm 3 .

| The SOC stock estimation for LUCAS 2018 samples
The BD fine and coarse volumefraction values calculated here can be used to estimate SOC stocks from the SOC content measures of the LUCAS 2018 dataset using (Poeplau et al., 2017):  We estimated the SOC stock values for the LUCAS 2018 samples using Equation 12, and we estimated the error that is introduced when the mass fraction of coarse elements is used instead of the volume fraction.

| The LUCAS 2018 BD fine earth dataset
We calculated BD fine and coarse volumefraction at the 0-20 cm depth for the selected LUCAS 2018 sites from coarse massfraction and BD sample data (Equations 6 and 11), and we classified sites based on their land cover. The statistics of the BD sample and the associated BD fine per land cover class is provided in Table 1. The statistics of the coarse massfraction and the associated coarse volumefraction per land cover class are provided in Table 2.
The most abundant classes were cropland, grassland and woodland soils, which accounted for 41%, 21%, and 30%, respectively, of the total number of selected sites (n = 6059); 'bareland' and 'shrubland' accounted for 3% of the sites each, whereas 'artificial land' accounted for <1%. Only six samples were classified as 'wetland'. The number of sites per land cover and per NUTS region is provided in Table S1.
We calculated the ratio BD fine =BD sample for all samples, and we found an average ratio of 90% with relatively small standard deviation (9%). The map of the values of BD fine =BD sample is shown in Figure 1. We found that BD fine =BD sample is lower in south-western Europe, Romania and Scandinavia. As a consequence, special attention should be paid when dealing with BD in these regions, because correcting for the volume of coarse fragments is particularly important.
Among land covers, we found that shrublands and woodlands had the lowest average ratios (85% and 88%, respectively), and thus are the most affected by the coarse fraction, while croplands and barelands had the highest average ratios (93% and 90%, respectively) and thus are less affected by the coarse fraction.
We calculated the ratio coarse volumefraction =coarse massfraction for all samples, and we found an average ratio of 41% with 13% standard deviation. The map of the values of coarse volumefraction =coarse massfraction is shown in Figure 2. Among land covers, woodlands had the lowest average ratio (31%). This is likely due to the lower median BD fine (Table 1) compared with the other land covers; therefore, the ratio ρ coarse BD fine is higher compared with the other land covers, so that the difference between the coarse volume fraction and the coarse mass fraction is higher (Equation 11). Consistently, we found the lowest values of coarse volumefraction =coarse massfraction in Scandinavia and Slovenia (Figure 2), where there are a large number of woodlands (Table S1).
3.2 | The SOC stock estimation for LUCAS 2018 samples using the new LUCAS 2018 BD fine earth dataset Figure 3a shows the coarse volume fraction as a function of the coarse mass fraction for varying BD fine values. Since BD fine < ρ coarse , we systematically find coarse volumefraction < coarse massfraction (Equation 11). The difference between the coarse volumefraction and the coarse massfraction is zero at the extremes (0% or 100% of coarse fragments) and is maximal at intermediate values. Additionally, the difference increases with decreasing BD fine because of the increasing difference in density between the fine earth fraction and the coarse fragments fraction (higher ρ coarse BD fine in Equation 11). Since coarse volumefraction < coarse massfraction , correcting SOC stock estimations by the coarse massfraction instead of the coarse volumefraction (Equation 12) causes an underestimation of the stocks (Figure 3b). The error is the most important for intermediate coarse fragment fractions and for soils with the lowest BD fine . Figure 3c, d shows the estimated absolute and relative error in SOC stock caused by using the coarse massfraction instead of the coarse volumefraction for the LUCAS sites. One data point with missing SOC content was discarded and data points having SOC content equal to zero (i.e., SOC content under the detection limit) were discarded in the estimation of the relative error in SOC stock (five points). Errors up to more than 80 Mg/ha are observed for sites with SOC stock > 100 Mg/ha (Figure 3c), and errors up to 80% are observed for sites with SOC stock < 10 Mg/ha (Figure 3d). Over the whole dataset, the median error is À2.57 Mg/ha (std = 6.67 Mg/ha) and the median relative error is À6.74% (std = 9.30%).  Table 3 reports the statistics of the estimated SOC stock errors per land cover. We found that the largest errors are expected in the SOC stock estimation of woodlands and shrublands. This is consistent with the fact that these land covers have the lowest BD fine =BD sample ratios and that woodlands have the lowest coarse volumefraction =coarse massfraction ratios.

| Limitations of the approach
The BD fine values provided in this study, and henceforth the SOC stock estimation, are approximate due to some limitations.
First, we rely on BD measures performed in the LUCAS 2018 campaign and on coarse fragments, measurements performed in the previous LUCAS campaigns. These may have changed slightly due to soil management. Second, the coarse fragments measures do not take into account large rocks that, in some land uses (and especially pastures), may be predominant. Third, we use an approximate and constant value for the coarse fragments density, as it is not measured in LUCAS. Finally, there are additional limitations for BD measurement accuracy generally related to the time of sampling and the agricultural management (e.g., tillage, crops in place, irrigation facilities). Both features can be controlled in a field experiment but are hard to take into account in a continental-scale sampling campaign.
In arable lands, BD is subjected to inter-annual variability due to sowing, root systems dynamics, and wetting and drying cycles (Moreira et al., 2016). Among the direct and indirect sources of variations in the EU, tillage, agricultural machinery involved in soil management, and trampling by animals can alter BD over the year (Colombi et al., 2018;Franzluebbers et al., 1995;Schillaci et al., 2021) with respect to undisturbed soil. Compaction might persist for years after disturbance (Bondi et al., 2021).
Nonetheless, the very large number of data points (>6000) should ensure that noise generated from sampling averages out over the whole dataset and can be assumed to have little impact on overall statistics.

| CONCLUSIONS
This work contributes to the EUSO and ESDAC 2.0, which have the main mission of timely delivery of the LUCAS soil monitoring data and assessing the status of European soils. Due to the policy interest in accurately assessing SOC stocks at the pan-European level, reliable BD measures that account for the presence of coarse fragments are needed. Such data can also be used to train predictive models at continental scale.
We provided a framework to better estimate the stocks of elements in soils by considering the coarse volumefraction to avoid underestimation. We provided the example of SOC stocks estimation, but the same applies for the estimation of stocks of any element, for example, nitrogen or phosphorus.
Notably, we showed that when SOC stock is computed with coarse mass fraction, the error compared with the computation by volume may vary depending on the SOC and coarse mass fraction. This may imply a SOC stock underestimation for European soils, especially in woodlands and shrublands. With remote sensing becoming higher resolution and more affordable, we hope that T A B L E 3 Statistics of estimated error in SOC stock generated when correcting by the mass fraction of coarse fragments instead of the volume fraction of coarse fragments per land cover. #: number of samples, std: standard deviation, Q1: first quantile, Q2: median, Q3: third quantile. Data points with missing SOC content measure were discarded. Data points having SOC content equal to zero (under the detection limit) were discarded in the relative stock error calculation. Values are approximated to the nearest integer. in the near future sampling can be deployed dynamically and that it will be possible to establish how many samples have been taken on arable land with altered soil structure due to management.
To summarize the contribution of this data article, we provide 6059 new BD fine and coarse volumefraction estimations for LUCAS 2018 topsoils. The data are publicly available in ESDAC 2.0 under LUCAS 2018 TOPSOIL data (https://esdac.jrc.ec.europa.eu/content/lucas-2018topsoil-data). This article wants to promote the use of good practice for data sharing and discoverability.