Soil health cluster analysis based on national monitoring of soil indicators

A major challenge in soil science is to monitor and understand the state and change of soils at a national scale to inform decision making and policy. To address this, there is a need to identify key parameters for soil health and function and determine how they relate to other parameters, including traditional soil surveys. Here we present a national‐scale dataset of topsoil sampled as part of a wider agri‐environment monitoring scheme in Wales, UK. Over 1,350 topsoils (0–15 cm) were sampled across a very wide range of habitats and a range of physical, chemical and biological soil quality indicators were measured. We show consistent differences in soil physicochemical properties across habitat types, with carbon decreasing and pH increasing across the habitat productivity gradient from bogs through woodlands and grasslands to arable systems. The soils within our dataset are largely within the limits identified as important for supporting habitat function, with the exception of excessive phosphate levels in mesotrophic grassland. Cluster detection methods identified four soil functional classes based on measured topsoil properties, which were more related to habitat type than the genesis‐based soil classification from soil maps. These soil functional classes can be interpreted as phenoforms within the soil genoforms found by traditional soil classification. This shows the importance of land‐use management in determining the soil health and functional capacity of soils. Our work provides an account of the current state of soil health in Wales, its relationship to soil function and a baseline for future monitoring to track changes against agri‐environment and other policy targets.

• There was a strong gradient in soil carbon and pH, with other variables being correlated with these • Mesotrophic grassland sites had phosphate levels above the identified limit for good functioning • Soil classes from topsoil properties were more related to land use than soil map classifications bulk density, carbon, cluster analysis, land use, nitrogen, pH, phenoform, phosphorus, soil health, texture

| INTRODUCTION
Soils underpin human existence through food, feed, fibre and timber production, as well as through earth system functions that support the delivery of other ecosystem services. Soil degradation affects 33% of all land globally according to the Intergovernmental Technical Panel on Soils (ITPS) (FAO & ITPS, 2015), and "52 % of the land used for agriculture is moderately or severely affected by soil degradation" as reported in Goal 15 of the United Nations (UN) sustainable development goals. In 2015, the first UN ITPS proposed four urgent actions to tackle and reverse degradation. The fourth was the development of robust soil monitoring systems to determine the current state and trend of soil health. Soil monitoring has become increasingly important in recent years, as nutrient loss, erosion and land-use change have implications not just for agriculture but for human activities as a whole. Land-use change impacts heavily upon soil function (FAO & ITPS, 2015), making integrated surveys for both soils and land management particularly important for understanding the impacts on land-use and climate change. The measurements we report here provide a baseline for the continuing monitoring of soil health and directly align with previous monitoring, allowing greater power to detect anthropogenic impacts on soil health. Traditionally, soil genesis and development studies have focused on processes occurring on the centennial to millennial timescales (Walker & Syers, 1976). However, there is an increasing recognition of the importance of sub-decadal changes in response to land-use change, pollution and climate drivers (Varallyay, 1990). This in turn is leading to a greater recognition of the importance of soil change and determining the speed of this change (Richter Jr & Markewitz, 2001;Tugel et al., 2005), and, perhaps more importantly, its potential impact on earth system function (Amundson et al., 2015;Schmidt et al., 2011). This shift in thinking has led to a difficulty in integrating traditional methods and results in soil science, which emphasize soil development and classification, with more recent needs for measuring and interpreting change in soil function, which recognize the more urgent need for evidence and action. In addition to traditional pedogenic-based classification (e.g., taxonomy), several approaches to bring together soil classification based upon genesis trajectories and results based on soil functional properties have been proposed. These include the Food and Agriculture Organization (FAO) topsoil classification (Broll et al., 2006), soil varieties within the Genetic Soil Classification of China (Shi et al., 2010), and the genoformphenoform concept (Droogers & Bouma, 1997;Rossiter & Bouma, 2018). Under the latter, soil classifications are seen as genoforms, which are time-invariant at human timescales (e.g., climate, long-term organisms, or land cover, relief and parent material acting through time). Soils that are sufficiently different within a genoform, and substantially affect soil function and are persistent over time, are classed as phenoforms (e.g., managed properties known to be important in soil function, such as pH and organic carbon). Genoforms act as fundamental controls on soil phenoforms and their function. This enables linkages between soil maps and function to be clearly expressed.
Soil functions are inherent capabilities of the soil that include biomass and food production, maintaining soil biodiversity, carbon and nutrient sequestration, water filtration and transformation, landscape and heritage, and being a source of raw materials (Blum, 2005). In order to track changes in soil functions, functional properties must be defined, which are required for monitoring at the national scale and need to be scalable to large areas and representative of functions across a variety of landscapes (Bünemann et al., 2018). This set of functional properties together represent a way to assess soil health. Here, we define soil functional properties as those that can be managed in a habitat-specific manner and are associated with the above functions. Therefore, we include carbon, pH, bulk density, nitrogen and phosphorus (Van Alphen & Stoorvogel, 2000). Soil carbon, pH, water content and bulk density are the most commonly proposed indicators for soil function due to their impacts on a wide range of soil functions (Bünemann et al., 2018). Bulk density, soil texture and associated water-related properties have been considered to be key indicators for monitoring of physical soil health (Corstanje et al., 2017). Soil carbon and nitrogen are key determinants of various soil functions, including greenhouse gas emissions, biomass production and influencing biological communities, but their exact impacts are often hard to evaluate (Gärdenäs et al., 2011). Other soil properties that have been found to be important in determining soil functions may be system dependent; for example, electrical conductivity and salinity have a large influence on soil functions when at the high levels that are more common in arid or intense arable systems, but are less important in other soil systems.
Wales, the location of our study, has recognized the role of soil in supporting wider ecosystem functions by inclusion of soil carbon as a key sustainability indicator within domestic legislation (Well-being of Future Generations (Wales) Act 2015). As awareness of the role of soils in supporting key ecosystem functions has increased, programmes to monitor and promote soil management have been put in place in various countries (e.g., Orgiazzi, Ballabio, Panagos, Jones, & Fernández-Ugalde, 2018), and in Wales this is integrated within the Glastir agri-environment land management scheme (Rose, 2011). In order to achieve these aims, current soils data are required to monitor changes in soil functions in response to wider ecosystem changes and their downstream effects. Data on soil properties that underlie health and function need to be collected using methods that are transferrable across the range of soils within Wales, but also the UK, Europe and globally, so comparisons can be made at large scales (Ribeiro, Batjes, Leenaars, Van Oostrum, & De Jesus, 2015). Frequency of data collection needs to be sufficient to detect changes within a politically relevant time period to allow for adaptive change of current policies as well as slower changes. The Glastir Monitoring and Evaluation Programme (GMEP) scheme meets these criteria in that it collects data on soil as part of an integrated monitoring programme covering vegetation, soil and water properties using a robust soil sampling methodology that has been used successfully across the variety of soils in Wales (Emmett et al., 2014). The GMEP scheme uses a methodology used in previous surveys in 1978, 1998 and 2007 to also allow for links to historical datasets. The GMEP soil measurements seek to address the need for data to understand the soil state and change at a national scale in order to inform policy.
Two approaches are commonly used to monitor longterm changes in soil properties: (a) localized monitoring of change in response to modifications of soil treatment, often in the form of field-scale manipulation experiments (Jenkinson & Rayner, 1977), and (b) large-scale "soil quality" surveys designed to inform land use and policy (Tóth, Jones, & Montanarella, 2013). Our approach is unique and differs from these in that national soil change and change in areas subject to management interventions are both measured through the same survey design. This enables the evaluation of land-management interventions for policy goals. The survey design is based on a stratified random approach developed for a Great Britain-wide integrated monitoring programme, the UK Centre for Ecology & Hydrology (CEH) Countryside Survey (Carey et al., 2008); an example map of site selection is shown in Figure 1. The soil monitoring programme also includes measurements not reported here relating to factors not routinely measured in large-scale soil surveys, such as a holistic evaluation of soil biodiversity . In addition, the survey allows for direct comparison between soil properties and aboveground factors, such as land-use change and plant species composition, as well as stream-water quality, due to the soil and aboveground surveys being co-located and adjacent streams and ponds being sampled. Here, we present results from the first iteration of this monitoring programme, a survey of topsoil (0-15 cm) health across Wales. We use these data to identify clusters of soils with similar topsoil properties and compare these classes with previously mapped soil groups. Our objectives are: 1 To present the topsoil results of a sub-decadal rolling agri-environment monitoring programme by habitat type 2 To determine if pH, Olsen P and bulk density values are within the nationally determined thresholds for habitat support 3 To evaluate the relationships between topsoil functional properties 4 To classify soils based on topsoil properties and compare these classes to land use and traditional soil classification methods 2 | METHODS

| Field measurement programme
Topsoil measurements were conducted through a 4-year field survey of 300 1-km squares across Wales (Figure 1), half of which are in areas prioritized by the Glastir agrienvironment scheme to determine the impact of land management interventions. The 1-km squares were selected at random from 26 land classes in proportion to their extent following the methodology of the UK CEH Countryside Survey (Carey et al., 2008;Reynolds et al., 2013), ensuring good coverage of the Welsh landscape. The initial survey took place over the summers of 2013 to 2016, and it is these results we present here. Each year, 75 squares were monitored, with each square having five soil sampling sites, each randomly located within a segment of the square. The soil sampling locations are centrally located within a 200-m 2 square quadrat that has a corresponding vegetation survey and habitat assigned by the surveyors according to the UK Biodiversity Action Plan broad habitat classification (Jackson, 2000). The soil cores for physicochemical analysis were taken with a corer of 5 cm diameter down to 15-cm depth after removal of vegetation and removal of any loose litter. The major soil group for each site was taken from the UK National Soil Map of England and Wales (Proctor, Siddons, Jones, Bellamy, & Keay, 1998).
Sites were selected by a random stratified sampling method, with half the squares being selected to provide a representative sample of the major land classes in Wales, whilst the remaining half were weighted towards habitats of particular interest. For the latter, each 1-km square had probability of being selected proportional to the score assigned to it under the Glastir Advanced Scheme by the Welsh Government. Models were used to estimate expected future Glastir scheme outcomes so that adjustments can be made to match Welsh Government priorities (climate change mitigation and water resources in years one and two) and scheme impact can be maximized. The national monitoring programme in Wales has evolved from the Countryside Survey soil sampling approach and methodology . In total there are: 20 supralittoral sediment sites, 39 arable sites, 388 improved grassland sites, 300 neutral grassland sites, 205 acid grassland sites, 79 broadleaf sites, 84 conifer sites, 86 heathland sites, 41 bracken sites, 53 fen and other sites, and 92 bog sites. Improved grassland is composed of fast-growing grasses typically managed as pasture or for silage production with the addition of fertiliser and/or lime. Neutral grasslands are usually found on soils with pH 4.5 to 6.5 and lack plants with strong preference for base-rich or acid soils. Acid grassland is characterized by plants with strong preference for acidic soils. Of the 1,387 sites, 1,353 had complete measurements for pH, carbon, nitrogen, total phosphorus and bulk density.

| Laboratory methods
The analysis of soil variables was performed using the methods employed in the Countryside Survey . In addition, soil surface water repellency was measured using the water drop penetration time method as described in Seaton et al. (2019). Details of the methodology are presented within the supporting information, and the full dataset is available from the Environmental Information Data Centre (EIDC) .

| Statistics
The differences in soil physicochemical properties by habitat were examined by providing summary statistics by habitat, counts of number of sites outside nationally determined threshold levels per habitat type, plotting F I G U R E 1 Map of Wales and the locations of the 300 individual survey squares. Locations are randomly shifted to any point on land within 10 km of the original location to ensure data confidentiality using the ggplot2 and egg packages (Auguie, 2019;Wickham, 2009) and mixed-effect models using the nlme package (Pinheiro, Bates, DebRoy, Sarkar, & R Core Team, 2019). Mixed-effect models were constructed with the square identity used as a random effect to account for spatial autocorrelation. Carbon, C:N, nitrogen, total phosphorus, Olsen P, rock volume and electrical conductivity were log transformed before modelling. Posthoc comparison of habitats was performed using the emmeans package with Tukey adjustment of p values for significance (Lenth, 2020). The relationships between the different soil properties were examined using Spearman rank correlations. Classification of the soils was undertaken using cluster analysis upon the soil properties considered to affect soil functions, including pH, bulk density, carbon concentration, water content, soil surface water repellency and total nitrogen (N). Soil properties such as total and available phosphorus and electrical conductivity are considered to affect soil function but had such low variation in our dataset that they were not included in this analysis. Soil water repellency was log 10 -transformed before inclusion. The clusters were fit using hierarchical clustering with Wardʼs criterion using the hclust function in R and four clusters were selected as the most appropriate divide based on the hierarchical tree (Murtagh & Legendre, 2011). The correlation of the clusters with the habitat groups was calculated using the χ 2 test and A r a b l e I m p r o v e d g r a s s l a n d N e u t r a l g r a s s l a n d A c i d g r a s s l a n d A r a b l e I m p r o v e d g r a s s l a n d N e u t r a l g r a s s l a n d A c i d g r a s s l a n d F I G U R E 2 Differences in soil pH (a), soil carbon concentration (b) and C: N ratio (c) across the range of habitats found in our study across Wales. Habitats are coloured by which aggregated habitat group they belong to and arranged in decreasing plant productivity order. The line bisecting each box represents the median value, with the box extending to the first and third quartiles of the data. The whiskers extend to the furthest values no more than 1.5 times the interquartile range.
Outliers are plotted individually.
Habitats that are significantly different from each other are indicated by differing letters below the boxplots (at p < .05) the strength of the correlation presented using Crámerʼs V statistic. All statistical analyses and graphing were performed in R version 3.6.1 (R Core Team, 2019).

| Soil properties across habitat types
Topsoil pH, carbon and nitrogen concentration vary across the different habitats found in Wales (Figure 2). Arable sites, improved grassland and neutral grassland tend to have the highest pH, lowest carbon concentration and lowest C:N ratios of the habitat types. The majority of all other habitat types have acidic soils (Figure 2a), with fen habitats having a slightly higher pH more similar to neutral grassland than the other habitats. Bog is an important carbon store, with a median carbon concentration of 490 g C kg −1 (carbon stock~6 kg C m −2 ), with acid grassland, coniferous woodland, heathland and fen having large ranges in carbon concentration and some sites having around 500 g C kg −1 (carbon stock >15 kg C m −2 ) (Figure 2b). C:N ratios are generally high across the different habitats, particularly in the high carbon habitats (Figure 2). Topsoil total nitrogen follows a similar variation across habitats to soil carbon, although total phosphorus and Olsen P show limited variation with habitat ( Figure S1). Bulk density varies considerably across habitat types, being highest in arable sites followed by improved and neutral grasslands and lowest in bogs ( Figure S2). Rock volume of soil and electrical conductivity show limited variation across habitat types ( Figure S2).
The broad habitats identified by the surveyors were aggregated into four habitat groups (improved land, neutral land, upland and woodland) for ease of interpretation. The ranges of carbon concentration, pH, nitrogen and phosphorus for the habitat groups are presented in Table 1. We do not present results for Olsen P in upland or woodland sites due to its methodological unreliability within low pH soils .
The soils were generally highest in silt-sized and sand-sized particles (Figure 3), with the majority being silty clay loam (n = 284) or clay loam (n = 232). There were also 145 sandy silty loams, 69 sandy loams, 19 silty T A B L E 1 Topsoil chemical properties: means ± SD, median, minimum and maximum, carbon concentration estimated from loss-onignition (LOI). Phosphorus measured as total phosphorus; Olsen-P results are only presented for improved land and neutral grassland clays, 15 silty loams, six loamy sands, six sands, four clays and one sandy clay loam. As the soil texture method involves organic matter removal prior to measurement it was only carried out on samples with lower loss-onignition carbon (LOI < 50%), so the carbon-rich soils are not included in these statistics.

| Thresholds
The majority of our sites are within the pH limits used as a national guideline for representing good support for the ecological habitat and biodiversity within specific habitat types (Bhogal, Boucard, Chambers, Nicholson, & Parkinson, 2008). We compare this to our new analysis of the Countryside Survey topsoil data, which compared the Welsh sites to the same thresholds (Table S1, Reynolds et al., 2013). Within the sites with mesotrophic grassland plant communities (i.e., improved and neutral grasslands), there are only 6% of sites that are outside the recommended pH range of 5-7 (Table 2). Two thirds of these 39 sites are deemed too acidic. This is considerably fewer mesotrophic grassland sites than have been identified as being too acidic in Wales in previous surveys such as the Countryside Survey (Table S1). However, in sites with acid grassland plant communities, 26% of our sites have pH above 5, which is considered to reduce their ability to support their distinct ecological communities (Bhogal et al., 2008). The previous Countryside Survey sites located in Wales found that the proportion of acid grasslands with pH above 5 increased over time from 1978 through 1998 to 2007 (Table S1). Countryside Survey soils in 2007 showed that 24% of acid grasslands had pH above 5, which is comparable to our result. A negligible proportion of our sites had bulk density above the identified threshold; however, the reliability of this threshold of bulk density as an indicator of soil status has yet to be fully tested due to a lack of data (Bhogal et al., 2008). Within mesotrophic grassland the extractable phosphorus (Olsen-P) was higher than the threshold for habitat support in three-quarters of the sites, which is similar to previous surveys ( Table 2, Table S1).

| Relationships between soil variables
Soils across Wales show that soil carbon concentration, bulk density and total nitrogen are highly correlated with each other. The relationship between carbon concentration and bulk density follows the distinctive curved shape found in previous studies of UK soils (Figure 4a) Howard et al., 1995). Total nitrogen follows a positive linear relationship with carbon at low concentrations of carbon, with a gradual levelling off and increasing variance at high carbon concentrations (R 2 = 0.87, Figure 4e). High carbon content soils are found solely in conjunction with low pH (Figure 4c). Plotting the Spearman correlations as a network shows the strong inner cluster of intercorrelated carbon, bulk density and nitrogen, which change in tandem across our sites (correlations~±0.9, Figure 5). Highly correlated with these three are pH, water content and soil water repellency. The rock content of the soil, electrical conductivity and total phosphorus are poorly correlated with the other soil parameters and situated on the edge of the diagram. For the exact correlation values see Table S2.

| Alternative soil classifications
Using the key soil parameters identified in the previous section (carbon concentration, total nitrogen, bulk The clay, silt and sand percentages of a subset of the soils (n = 781) plotted on a ternary diagram T A B L E 2 Number of sites above the UK national guidelines set by the environment agency. For Olsen-P this is 10 mg/L for mesotrophic grassland; no results are presented for acid grassland and heathland. For pH this is <5 and >7 for mesotrophic grassland, >5 for acid grassland and heathland. For bulk density this is above 1.3 g/cm 3 for mesotrophic grassland and 1.0-1.3 g/cm 3 for acid grassland and heath

Habitat
Olsen-P pH density, pH, water content and water repellency) we placed our soils into different categories. The cluster dendrogram from the k-means method of clustering usefully organized the hierarchical patterns of similarity in the dataset. This allowed the most ecologically informative clusters to be identified, striking a compromise between too few with too much internal variance versus too many similar groups, resulting in four approximately equally sized categories ( Figure 6). We plotted these soil categories against the properties used to create these clusters and named the categories as organic, organo-mineral, acid mineral and neutral mineral soils ( Figure S3, Table S3).
The classification of soils into our categories showed a stronger relationship with the aboveground habitat than the major soil group within the mapped classification by genesis did (Figure 7). Both soil classification systems were significantly associated with the aggregated habitat group (χ 2 test, p < .001); however, the relationship between the soil topsoil properties classification and habitat was stronger than the relationship between the soil genesis classification and habitat (Crámerʼs V was 0.455 and 0.301, respectively). The results for broad habitat were also significant and showed the same pattern of strength. The topsoil properties classification strongly separated out the bog, which was found only in the F I G U R E 4 The major soil parameters plotted against each other and coloured by habitat group (n = 1,367, 1,363, 1,367, 1,362 and 1,364 for panels (a) to (e), respectively) organic class, and the arable site, which was found almost solely on neutral mineral soils. All other habitats showed a definite trend with the topsoil properties classification. Improved and neutral habitats were more likely to be associated with acid and neutral mineral soils, or brown soils in the case of the mapped soil classes. There are differences in the proportions of topsoil properties clusters per each mapped soil unit (χ 2 test, p < .001, Crámerʼs V = 0.356), but every mapped classification had at least one example of every topsoil functional cluster, with limited differences in proportions across the three most numerous mapped soil classes ( Figure S4).

| Key soil parameters status and correlations
The range and distribution of soil physicochemical properties found within this survey are in agreement with previous national-scale surveys of UK soils (Baxter, Oliver, & Archer, 2006;Bellamy, Loveland, Bradley, Lark, & Kirk, 2005;Reynolds et al., 2013). The trend in carbon and pH with habitat showed that arable and improved grassland habitats had the lowest carbon concentration and acidity, with these increasing in the lower productivity habitats such as bogs and heathland. Wales contains many carbon-rich, low-pH soils, which are not always included in other surveys due to their focus on soils of high agricultural production (e.g. Baxter et al., 2006). However, the improved lands included within our survey had pH levels consistent with previous studies of agricultural lands within Wales (Baxter et al., 2006;Reynolds et al., 2013). Compared to the rest of Europe, Welsh soils have on average higher carbon concentration and lower pH; this is also true when only comparing them with the soils from the Atlantic climatic region (Tóth et al., 2013). This may be related to the general dominance of an acid geology and the high precipitation, as globally there is lower pH in areas with greater precipitation, which is thought to be linked to acidity and slower decomposition, thus enabling a build-up of soil organic matter (Slessarev et al., 2016). Our soil texture results also support evidence from the National Soil Survey that Wales is lacking in finer-grained, clay mineral soils compared to the rest of the Atlantic region of Europe (Tóth et al., 2013). All of these properties will contribute to the generally low productivity of many Welsh soils and infrequent presence of arable farming systems.
The relationships between the different soil physicochemical variables are consistent with those found previously in the UK Countryside Survey, especially the strong correlation between carbon concentration and pH and other variables such as total nitrogen and bulk density (Reynolds et al., 2013). The distinctive curved negative association of carbon concentration with bulk density pH C N BD Wtr SWR P EC Rck F I G U R E 5 The Spearmanʼs rank correlations between the variables plotted as a network. Each circle (node) is a variable and the lines between circles represent the correlation between those two variables across the entire network. The width of the line is proportional to the strength of the correlation and the lines are coloured with blue for positive correlations and red for negative correlations. The layout of the network is selected by an algorithm that attempts to put strongly related variables closer together. The node labels correspond to: BD, bulk density (log); C, carbon concentration; EC, electrical conductivity (log); N, total nitrogen; P, total phosphorus; pH; POI, Olsen-P; Rck, rock volume in soil; SWR, soil water repellency (log); Wtr, volumetric water content  Figure 7. From left to right: group 1 (308 members, orange: improved grassland), group 2 (280 members, green: woodland), group 3 (350 members, pink: neutral grassland) and group 4 (437 members, blue: upland) has been found by many studies across a variety of climatic zones Howard et al., 1995;Périé & Ouimet, 2011). There was limited correlation of total phosphorus, Olsen-P phosphate or electrical conductivity with the other variables in our dataset, which suggests phosphorus supply is not primarily linked to organic matter formation but rather the composition of the soil parent material and potentially in some cases the external supply of phosphorus from fertilisers.
The comparison of our soils to nationally set thresholds enables us to increase our understanding of soil health across Wales in a domain-specific manner. Some thresholds are relatively well established, such as those for pH and extractable phosphorus (i.e., Olsen-P), which have been identified for different environmental interactions and habitat support (Bhogal et al., 2008). Other indicators have been proposed, such as bulk density, soil carbon and C:N ratio, but often there is limited evidence and/or consistency across ecosystems in the impact of these (Bhogal et al., 2008). Bulk density, together with clay content, has been proposed as a soil quality indicator for British soils in relation to trends over time, rather than passing a pre-identified threshold (Corstanje et al., 2017). We have limited sites at the higher levels of bulk density, which makes it difficult to evaluate the threshold value, and overall there was little correlation between bulk density and soil biological indicators such as total mesofauna (George et al., 2017). Bulk density within our data strongly correlated with soil carbon, which may suggest that in these types of soil systems bulk density and carbon represent the same aspects of soil health. Instead of a single threshold, it has been suggested that evidence of decreasing soil carbon acts as a trigger value, in particular due to its relevance to carbon storage and biogeochemical cycling. The continuation of this survey in the coming years will allow clear identification of any habitats that may be losing carbon and thus should be targeted for land management interventions. Past surveys for Wales using the same methodology and some common locations in the Countryside Survey did not identify any consistent trends in C concentration or density between 1978between , 1998between and 2007between (Reynolds et al., 2013. Other survey and modelling approaches have suggested soils for the UK are on average losing carbon (Bellamy et al., 2005;Jenkinson, Adams, & Wild, 1991;Jones et al., 2005) or are remaining stable (Reynolds et al., 2013;Smith et al., 2005). However, trends appear to be highly specific to land-use types (e.g. soil C loss in arable soils but gains in woodland soils) (Reynolds et al., 2013) and thus country-level trends perhaps mask important trends linked to specific management practices within land-use type. It is important to note also that this survey does not measure changes in subsoil carbon, which is critical for carbon storage and likely to be less influenced by land use than topsoil carbon, limiting the inferences that can be made about the overall soil carbon stock and changes (Harrison, Footen, & Strahm, 2011;Simo, Schulte, OʼSullivan, & Creamer, 2019).
We have found that three-quarters of our mesotrophic grassland sites with Olsen-P were above the trigger value related to habitat support, which is similar to the 60% and 90% of mesotrophic sites that we found to be above the trigger value in the Countryside Survey in 2007 and 1998, respectively. Globally, phosphate decreases in grasslands have been predicted from model data (Sattari, Bouwman 2016) and have been previously reported in Wales from Countryside Survey data (DeLuca et al., 2015;Reynolds et al., 2013). This is linked to the 60% reduction in use of P fertilisers in the UK from the 1980s to 2010, which has since stabilized (The British Survey of Fertiliser Practice, 2019), and could be expected to have an impact across the landscape even on unimproved land due to reduced transfer of phosphate by hydrological or atmospheric pathways. Reduced grazing could be decreasing the removal of phosphate by preferential grazing on phosphorus-enriched areas and thus reducing its diversion away from grazing sites (Schütz et al., 2006;Statistics for Wales, 2016). High levels of phosphate have been linked to lower plant diversity (Critchley et al., 2002;Michalcová, Gilbert, Lawson, Gowing, & Marrs, 2011). Moreover, elevated phosphate levels can persist in the soil for long periods and have lasting impacts on the plant communities that can establish at a site (Horrocks et al., 2016). The current proportion of sites that are outside the pH thresholds is comparable to the most recent surveys of the Welsh countryside by the Countryside Survey in 2007, which shows markedly less acidity overall than surveys within 1978 and 1998 (Reynolds et al., 2013). This can be interpreted as stabilization rather than continued recovery from historic acidification due to atmospheric acid deposition. This does not necessarily mean that Welsh soils are fully recovered from acidification, as there are some indications from model data that recovery from acidification is not yet complete (RoTAP, 2012). The stalling of recovery from acidification could indicate that the soils have entered into a lower pH stable state (Suding, Gross, & Houseman, 2004); thus, to enhance productivity some soils may require active remediation to return to pre-acidification pH. Similar results showing reductions in recovery from acidification are reported for woodland and other organic soils (Evans, Monteith, Reynolds, & Clark, 2008;Kirk, Bellamy, & Lark, 2010;Reynolds et al., 2013), and are attributed to vegetation uptake of base cations, nitrogen deposition or capture of acidic pollutants by the woodland canopy, which offsets SO 2 reductions. This does, however, raise the issue as to whether the assumption that a pH of less than 5 is required for habitat support in acid grassland and dwarf shrub heath will hold, as the impacts of anthropogenic acidification are reduced and soil pH values increase across the UK. The shifting baseline in soil pH may be altering our perception of what constitutes a good pH value for an acid grassland (Soga & Gaston, 2018). The thresholds for pH were established based largely on data from UK grasslands in the 1990s (Bhogal et al., 2008), which would represent sites that are in the process of recovery from intense acidification and therefore may not actually be similar to a true natural state. The pH trigger values for supporting metal retention and microbial function are actually contradictory to those suggested for supporting acid grassland and heathland habitats (<5 and >5, respectively), and recent results indicate microbial function may decrease below pH 5.5 rather than 5, exacerbating this difference (Jones, Cooledge, Hoyle, Griffiths, & Murphy, 2019). This shows the difficulties in designating appropriate boundaries when multiple functions and services are involved, especially when the different functions show differing responsiveness to change (Bhogal et al., 2008;Bünemann et al., 2018;Jarvis et al., 2019).

| Soil classification
The soil physicochemical clusters identified in this work have strong similarities with previous analysis of UK soils. Analysis of the soils collected as part of the Countryside Survey of the UK in 2007 found that there were three main clusters of soil physicochemical properties corresponding to mineral soils, organo-mineral soils and organic soils (Simfukwe et al., 2010). In our data we split the mineral soils into two groups; however, in other respects our classifications are similar. These results support the use of soil organic material in categorizing soils, as evidenced by the use of carbon classifications within multiple classification systems (Broll et al., 2006;Emmett et al., 2010).
Our analysis has shown that, as we hypothesized, the traditional soil classification methods, such as that used within the UK soil classification (Avery, 1980), are weakly correlated with differences in habitat type and land use, whereas those based on key topsoil manageable parameters are more strongly related. This is consistent with previous results showing that soil dissolved organic carbon is not well related to soil type in UK soils (Simfukwe, Hill, Emmett, & Jones, 2011). We have also found that there seems to be only a limited relationship between our identified topsoil class and the traditional classification, with the exception of peat soils. This suggests that there may be limited constraints from the soil genesis type upon the functional nature of the topsoil, indicating the importance of management decisions in determining soil function. The functional capacity of the subsoil, however, may be more constrained by the soil genesis type than land cover. Therefore, soil functions that are dominated by different soil horizons may have been influenced more or less strongly by the plant community versus the soil genesis type. One key limitation of this analysis is that the soil classification by genesis was taken from a map based largely on data collected from 1960 to 1970, which fails to capture changes in soil management and land use that have occurred since then. In addition, our survey locations were classified at the soil association rather than series level. Soil surveying to classify soil genesis is time consuming and labour intensive, often making funding of large-scale soil surveys unattractive. In practice, many key survey and modelling results are either based on previous soil mapping efforts or on topsoil sampling only, which is what we have compared here to land cover. Many ecosystem service maps use the soil classification maps, when actually soil function is more related to the plant communities and land-cover type.
The soil properties we have presented here are a selection of properties that are known to influence soil function and are both manageable and measurable at a national level. The soil properties that are often measured in the scientific literature to represent function (e.g., carbon mineralization rates) (Simfukwe et al., 2011) are usually difficult to scale up to large areas due to factors such as expense and limited generality across landscapes (Sanchez, Palm, & Buol, 2003). Many of these properties can also be only measured in a laboratory environment on highly processed soils, which means that they can fail to capture the conditions as they really exist, particularly the role of plants in regulating soil functioning (Carlyle, Nambiar, & Bligh, 1998;Oburger & Jones, 2009). Different functions can also respond differently to the soil properties considered here and even soil biodiversity can be represented by different aspects with different responses. For example, in our sites microbial diversity is highest in our habitats with high pH and low carbon , whereas mesofaunal abundance is highest in habitats with more intermediate pH and carbon (George et al., 2017). However, some properties, such as soil carbon, have been widely accepted to be indicators of soil function, influencing greenhouse gas emission, nutrient cycling, water filtration and biomass production, among others (Amundson et al., 2015;Bünemann et al., 2018;Environment Audit Committee, 2016;Rossiter & Bouma, 2018). It is these parameters -pH, carbon, nitrogen, bulk density and waterthat we have found to be pivotal in determining the topsoil property classes of our Welsh soils; therefore, we term them functional clusters.
The clusters we have found behave differently in their functional attributes, reflecting their different land management regimes. The key functional attributes of soil vary depending upon their pedogenic characteristics and the overlying land use. We see the classes proposed by our analysis as a way of reducing complexity to enable comparison of like-for-like, and consequently, we do not apply the principles of functionality derived from lowland arable soils to upland peatlands. This comparison of appropriate classifications is particularly relevant for determining policy at the national scale, when balancing the need for provision of multiple functions across a heterogeneous landscape. There is no way to tell within our data whether differences in areas targeted for agrienvironment interventions are due to the scheme or preexisting conditions, and thus we have not evaluated that here. However, the dataset presented here offers an understanding of the current state of soil health in Wales that can be used as a baseline for future surveying to evaluate the response of soil health and function to land management interventions. The differences in soil health and function across habitats we have found show the importance of land management to soil function.
There have been objections to the principle of classifying soils into strictly defined categories since the advent of soil classification systems (Webster, 1968). In response, many authors have chosen to use fuzzy mathematical methods to classify soils (Burrough, 1989;Mazaheri, Koppi, & McBratney, 1995;Stevenson, McNeill, & Hewitt, 2015). This can allow any given soil to belong to more than one class, potentially better capturing the range of soils between different classes than the artificially abrupt boundaries between classes in a hierarchical classification system. Soils generally exist on a continuum in trait space, exhibiting different characteristics across a variety of landscapes. They can also change over time and under different management practices, particularly those already at the edge of the categorization boundaries. Results such as ours, which find certain categories of soils based on their properties, should be interpreted within this context. Although categorization is a useful tool for informing management and monitoring, it cannot represent the full breadth and flexibility of soil types.
The clusters of soils we have identified can be aligned to the phenoform concept, where the phenoforms are the functional clusters, which can be nested within the genoforms (i.e., the mapped soil classes by genesis). However, we have found that the genoform poses no major constraint upon the types of phenoform that can develop there, which suggests the nested nature of the genoform-phenoform concept may be an unnecessary complication in practice, at least with respect to topsoil. One issue with the comparison of our results to the genoform-phenoform concept is that the phenoform definition considers only soil properties that are persistent and require substantial management change to alter (Rossiter & Bouma, 2018). The properties often identified as being key to functional classification, such as carbon and pH, are experiencing ongoing change and are the target of key initiatives such as the 4 per mille initiative, which aims to increase global soil organic matter stocks by 0.4% per year (Minasny et al., 2017). There is a conflict in the application of the phenoform concept that hinges on the identification of what constitutes "substantial" management. This conflict reaches its peak when considering changes over time. If we were now to find that the 4 per mille initiative was successful then this would constitute enough change to alter the phenoform of all of our soils. But if all change in tandem, as occurred with the recovery from acidification in UK soils (Reynolds et al., 2013), then new attempts to define phenoforms on the basis of cluster analysis of soil properties will not show these changes and find the same phenoforms again. It may be unlikely that different areas will respond in tandem to external changes due to differences in the application of these changes, the inherent differences in responsiveness of different habitats, and the non-linearities of change directions as indicated in the fundamentally different direction of soil carbon losses within different land-use types reported by Reynolds et al. (2013). However, the direction and magnitude of change within soils is a key constraint on the application of the phenoform concept that requires further investigation. The value of repeated soil monitoring in establishing any trends in the health and presence of phenoforms cannot be overstated, as soil health is dynamic at managementrelevant timescales.

| CONCLUSIONS
We present a national dataset, which provides a baseline for the survey of Welsh topsoils (0-15 cm), allowing for the quantification of the current health of the soil and enabling future surveys to track trends in these conditions. We show that there are consistent differences in soil properties across habitats. Few of our soils are outside established thresholds of pH and bulk density for ecosystem health, but high levels of phosphate in improved grasslands remain an issue. Several key soil properties, such as carbon, nitrogen and pH, are strongly correlated across our soils and can be used to create a classification of the soils. We propose that our conceptual classification of the topsoil is related to soil functionality, due to the known relationships between the key soil properties featured here and soil functions. Consequently, the functional classification of the topsoil developed in the present analysis is more related to land-use type than soil classes based on traditional methods.