Exploring relationships between land use intensity, habitat heterogeneity and biodiversity to identify and monitor areas of High Nature Value farming

Understanding how species richness is distributed across landscapes and which variables may be used as predictors is important for spatially targeting management interventions. This study uses finely resolved data over a large geographical area to explore relationships between land-use intensity, habitat heterogeneity and species richness of multiple taxa. It aims to identify surrogate landscape metrics, valid for a range of taxa, which can be used to map and monitor High Nature Value farmland (HNV). Results show that variation in species richness is distributed along two axes: land-use intensity and habitat heterogeneity. At low intensity land-use, species rich groups include wetland plants, plant habitat indicators, upland birds and rare invertebrates, whilst richness of other species groups (farmland birds, butterflies, bees) was associated with higher land-use intensity. Habitat heterogeneity (broadleaved woodland connectivity, hedgerows, habitat diversity) was positively related to species richness of many taxa, both generalists (plants, butterflies, bees) and specialists (rare birds, woodland birds, plants, butterflies). The results were used to create maps of HNV farmland. The proportion of semi-natural vegetation is a useful metric for identifying HNV type 1. HNV type 2 (defined as a mosaic of low-intensity habitats and structural elements) is more difficult to predict from surrogate variables, due to complex relationships between biodiversity and habitat heterogeneity and inadequacies of current remotely sensed data. This approach, using fine-scaled field survey data collected at regular intervals, in conjunction with remotely sensed data offers potential for extrapolating modelled results nationally, and importantly, can be used to assess change over time.


Introduction 1
Agriculture has been a major driver of global environmental change and unprecedented 2 biodiversity loss over the past century (Benton et al. 2003;Firbank et al. 2008;Strohbach et al. 3 2015). Agricultural intensification involves increases in external inputs (pesticide and fertilisers), 4 land-use change, increases in field sizes and fragmentation and loss of semi-natural habitats; all of 5 these have caused the decline of many different taxa (Billeter et al. 2008;Chamberlain et al. 2001;6 Robinson & Sutherland, 2002). However, agriculture is important for food production; croplands and 7 pastures cover 40% of the global land surface (Foley et al. 2005) and many species are dependent 8 upon agricultural habitats (Benton et al. 2003). Therefore, biodiversity protection globally depends 9 upon conservation in these human-dominated landscapes (Fahrig et al. 2011;Karp et al. 2003). 10 Evidence suggests that biodiversity can be increased by changing to low intensity land uses (Bignal & 11 McCracken 1996; Karp et al. 2003) or by changing landscape structure, e.g. increasing landscape 12 heterogeneity and connectivity (Stein et al. 2014;Benton et al. 2003;Steffan-Dewenter 2003). It may 13 be more difficult to take land out of production because of farmer livelihoods and requirements for 14 food (Fahrig et al. 2011). However, where agriculture depends on structural support payments, 15 reductions in these could drive abandonment on marginal land (Renwick et al. 2013). Low intensity 16 systems, for instance, semi-open habitats maintained through extensive grazing, are important for 17 many priority species (Lubos Halada et al. 2010;Woodhouse et al. 2005). 18 Landscape heterogeneity can moderate the negative effects of local land-use intensity 19 (Perovic 2015). Increased compositional heterogeneity (diversity of habitat types) represents more 20 niches which support more species, whilst configurational heterogeneity (number, size and 21 arrangement of habitat patches) (Fahrig et al. 2011, Perovic et al. 2015 increases the variability of 22 microclimatic conditions and provides breeding sites (Stein et al. 2014;Benton et al. 2003), whilst 23 increasing the ease with which species can move through the landscape and achieve viable 24 metapopulations (Lawton et al. 2010). However, high habitat heterogeneity can have negative 25 effects by increasing habitat fragmentation, at the expense of habitat specialists (Fahrig et al. 2011). 26 Agricultural landscapes vary widely in the degree of intensity of production and spatial 27 heterogeneity, and by land ownership, historical and cultural practices, topography and soil type 28 (Fahrig et al. 2011). To protect and maintain farmland biodiversity requires a framework for priority-29 setting. In Europe, the High Nature Value (HNV) farmland concept was introduced as 'areas in 30 Europe where agriculture is a major (usually the dominant) land use and … supports or is associated 31 with either a high species and habitat diversity, the presence of species of European concern or both' 32 (Andersen et al. 2003). Thus 'the preservation and development of HNV farming systems' is a 33 strategic priority for EU member states and contributes to targets for halting biodiversity loss by 34 2020, so subsidies are prioritised to HNV areas (Brunbjerg et al. 2016). These tend to be marginal for 35 farming with low productivity. They produce multiple ecosystem services such as carbon storage, 36 clean water, and aesthetic landscapes. 37 38 Three HNV types are broadly recognised (Paracchini et al. 2008): Type 1-farmland with a 39 high proportion of semi-natural vegetation; Type 2-farmland with a mosaic of low intensity 40 agriculture and natural and structural elements, e.g. field margins, hedgerows, scrub, small rivers; 41 Type 3-farmland supporting rare species or a high proportion of European or world populations (can 42 occur at small scales in an otherwise intensively managed landscape). 43 44 The assumptions underlying the HNV types 1 and 2 definitions, that high species richness is 45 associated with high habitat heterogeneity and low intensity land-use, are evidence-based (Stein et 46 al. 2014 (Fahrig et al. 2011), as a single species may not be a good predictor of 55 other species groups (Billeter et al. 2008;Firbank et al. 2008). 56 HNV farming is the only Common Agricultural Policy (CAP) impact indicator for which there 57 is no common methodology explicitly provided at the European union (EU) level. Each Member State 58 uses data and methodologies suited to their prevailing bio-physical characteristics and farming 59 systems, and based on the highest quality and most appropriate data available, including for 60 instance, landscape elements (hedgerows) and indicator species (particularly birds and plants) 61 (Klimek et al. 2014, Morelli et al. 2014, Brunbjerg et al. 2016). 62 There have been attempts to create a system for identifying HNV farmland consistently 63 across Europe using various approaches, including land cover, farming system, protected areas and 64 species (Andersen 2003;Beaufoy et al 1994). Most European-scale approaches lack the spatial and 65 temporal resolution necessary for national and regional application (Lomba et al. 2014). 66 Even at national and regional scales it can be difficult to obtain data at high resolution on 67 landscape elements, farming intensity, management practices (Strohbach et al. 2015) and 68 biodiversity. Coarser, spatially continuous, remotely sensed data may be available but do not 69 provide the detail of finely resolved data (Wood et al. 2018), for instance, small biotopes and 70 hedgerows cannot be easily detected by remote sensing, and data are not necessarily available at 71 the appropriate frequency to monitor change. Where biodiversity data are available they are often 72 sampled data such as the bird surveys carried out for common bird monitoring in the UK (Harris et 73 al. 2018) which cover selected sites but make it difficult to produce continuous maps (Strohbach et 74 al. 2015). 75 76 Here, we develop methods to integrate fine-scaled, sampled data (for biodiversity, 77 landscape heterogeneity and structure) with coarser, spatially continuous data from remote sensing 78 (Boyle et al. 2015, Klimek et al. 2014 to enable extrapolation outside of the sampled sites. This 79 study is at a national scale (Wales) and uses data collected as part of the monitoring project (GMEP; 80 Glastir Monitoring and Evaluation Project) designed to detect the impacts of the Glastir agri-81 environment scheme (the main scheme by which the Welsh Government pays for environmental 82 goods and services funded by the EU's Rural Development Programme (RDP)). 83 84 This study: i) explores the relationships between elements of land-use intensity, habitat 85 heterogeneity and species diversity (using a range of taxa) to support the use of metrics to identify 86 HNV types 1 and 2; ii) uses the results of those analyses to identify key explanatory variables that 87 could be used to scale up nationally from fine-scaled analysis of field survey samples and iii) maps 88 High To explore these relationships, we used data from the Glastir Monitoring and Evaluation programme 109 (GMEP). The methodology is based on that of Countryside Survey (Smart et al. 2003, Norton et al. 110 2012), with some methodological differences (Emmet et al. 2015). Over 4 years, 300 1 km squares 111 were sampled, half of these were based on a stratified random sample by land class (e.g. geology 112 and soils), and the other half a random sample weighted towards Welsh government priorities for 113 options within Glastir. Within each 1 km square, a series of measurements were taken. The metrics 114 used are outlined in Table 1 A series of up to 50 vegetation plots sampling different features were located within each 1 km 119 square (Smart et al. 2003). Linear features (watercourses, hedges and field boundaries) and areal 120 features (fields, unenclosed land and small semi-natural biotope patches) were sampled. Linear plots 121 were 1 m x 10 m laid out along a feature. Area plots were randomly placed (2 m x 2m), while a series 122 of targeted plots sample small habitat patches and habitats of conservation value. In each 123 vegetation plot, a list was made of all vascular plants and the more easily identifiable bryophytes. 124 Response variables calculated from the vegetation plot data for each 1 km square include: mean 125 number of total plant species per plot, mean number of ancient woodland indicator species per plot 126 (Kimberley et al. 2013), mean number of wetland species per plot and mean number of species 127 indicating high quality habitats. The latter was created from a list of plant indicator species taken 128 from the Common Standards Monitoring guidance for Sites of Special Scientific Interest (JNCC 129 website) and refined in consultation with the Botanical Society of the British Isles from a list of 130 axiophytes ('worthy' plants indicative of habitats of high conservation value). The mean number of 131 wetland species per plot was calculated using this list for wetland habitats only. The ancient 132 woodland indicators were identified in a separate list collated from discussions with woodland 133 experts. 134 135 Birds 136 The bird surveys were carried out by BTO. The survey protocol operated at the same spatial scale (1  137 km squares) as the national BTO/JNCC/RSPB Breeding Bird Survey (BBS), but involved more intensive 138 fieldwork in space and time (Emmett et al. 2015). The surveys consisted of four visits to each square, 139 equally spaced through mid-March to mid-July. On each visit, the surveyor walked a route that 140 passed within 50 m of all parts of the survey square to which access had been secured, taking up to 141 five hours. All birds seen or heard were recorded on high-resolution field maps using standard BTO 142 activity codes. Bird data were summarised to calculate the number of woodland bird species in a 1 143 km survey square (species-specific maxima across all four visits), and the same for farmland birds, 144 upland birds and rare birds. There are defined species and habitats of principal importance to 145 conservation in Wales that are known as 'Priority' or Section 7 species and habitats (Wales 146 Environment act) and the rare birds are taken from that list (A1). The woodland bird index and the 147 farmland bird index are well-established for reporting at national level in the UK and mainland 148 Europe (Gregory et al. 2008). 149 150 Pollinators 151 Butterfly Conservation organised the survey of pollinators focused on three main pollinator groups: 152 butterflies (Lepidoptera: Rhopalocera), bees (Hymenoptera: Apoidea) and hoverflies (Diptera: 153 Syrphidae). Butterflies were recorded to species level, whilst bees and hoverflies were recorded as 154 groups (A2) based on broad differences in morphological features associated with ecological 155 differences. Shannon diversity indices were calculated using the number of bee and hoverfly groups 156 recorded, to account for evenness. A 2 km transect route was taken through each 1 km square 157 (following the UK Butterfly Monitoring Scheme, Brereton et al. 2011), all butterflies within a 5 m box 158 are recorded while walking a fixed route at a steady pace under a set of pre-determined weather 159 conditions and at a set time of day (known as 'Pollard walks', Pollard 1977). Hoverfly and bee groups 160 were also counted simultaneously along the same transects. Pollinator metrics used in this analysis 161 include bee species diversity index, hoverfly diversity index, butterfly species richness, woodland 162 butterfly species richness and species richness of rare invertebrates (Section 7, Wales Environment 163 Act). 164 165 2.2.2 Habitat Heterogeneity 166 Habitat Diversity 167 Habitat areas (>20 m x 20 m) were mapped and classified in the GMEP field survey onto hand held 168 computers using the Broad and Priority Habitat classification (Jackson 2000). Shannon's diversity 169 index was calculated to take into account the number of Broad habitats and the dominance among 170 them (Firbank et al. 2008). 171 172 Habitat patch size 173 Mean area of habitat per 1 km square was calculated from field survey mapping data. 174 175 Linear features 176 Linear features (less than 5 m wide, minimum length 20 m) recorded include the length of managed 177 hedgerows, unmanaged lines of trees, streams and ditches in each 1 km square. 178 Connectivity of woodlands and wetlands 179 Habitat connectivity is a function of the number and size of habitat patches and how close together 180 they are; this was estimated from the habitat maps recorded by the field survey team. We 181 considered Euclidean distance (distance in metres between the edges of each habitat patch) and 182 least-cost methods, and used least-cost for fine-scaled data. Least-cost paths were calculated as a 183 function of the landscape occurring between two habitat patches, using expert judgement of the 184 ease of movement of a generic broadleaf woodland or wetland species to assign weightings to each 185 habitat (Jackson et al. 2013 The soils of Wales are mapped as part of the soil survey of England and Wales (Avery, 1980 (2017) used 98 soil associations taken from the 208 soil survey of England and Wales in an analysis to identify rare soils and to assess spatial patterns 209 (soil diversity) across Wales and these data were used here. Soil diversity is measured using the 210 Shannon diversity indices similarly to the calculation for habitat diversity (Maxwell et al. 2017 Generalised Additive Modelling (GAM) (Hastie & Tibshirani, 1990) in R (R core team 2017) was used 214 (with a Poisson distribution) to analyse interactions between species richness of biodiversity 215 indicators and explanatory variables (Table 1). Spatial autocorrelation (SAC) was tested by extracting 216 the model residuals and testing with Moran's I in R (using functions in the 'ape' library) (Dormann et 217 al. 2007). Results suggested that there was SAC for some variables (birds, butterflies, bees: p<0.001), 218 so we accounted for SAC by specifying a spatially explicit model for the residual structure with the 219 nlme package, which provides functions for spatial correlation structures (Dormann et al. 2007 Habitat diversity 242 Habitat diversity was calculated using the method described above but using the LCM2007 Broad 243 habitat classes rather than Field Survey data. 244 245 Woody Linear features 246 The percentage cover of woody vegetation was calculated using airborne radar data (NEXTMap®), 247 optical imagery from satellites and data from the National Forest Inventory. NextMap provides 248 canopy height information at 5 m x 5 m spatial resolution and this dataset was used to identify 'tall' 249 features in the landscape. Normalised Difference Vegetation Index (NDVI) imagery was used to 250 separate vegetated from non-vegetated areas. NDVI was derived using data from the Landsat 8 251 Operational Land Imager (OLI), calibrated to reflectance and masked to remove cloud and cloud 252 shadow. NDVI was calculated using: 253 254 Larger areas of woodland were supplemented by the National Forest Inventory 2013 dataset to 257 produce a woody features product with a binary (woody/non-woody) classification at 5 x 5 m spatial 258 resolution (Tebbs & Rowland, 2014). 259 260 Connectivity 261 Wales was divided into ~20,000 1 km squares for which area and location of broadleaf woodland 262 and wetland were assessed from LCM. Within the GMEP field survey squares the least-cost 263 connectivity metric was compared to the Euclidean distance metric and there was a high significant 264 correlation (r squared=0.95, p<0.001) so, for the all Wales dataset, Euclidean distances were used to 265 reduce processing time. The pairwise distances and size of each fragment were used to calculate the 266 probability of connectivity metric for each 1 km grid cell using Conefor software (Saura & Torné,  267 2009), as for the field data. 268 269

Land-use intensity 270
The percentages of semi-natural and improved land were calculated in the same way as above but 271 using remotely-sensed LCM2007 data rather than field survey data. 272 273

National Analysis 274
To compare the use of explanatory variables from fine-scaled field data with remotely sensed data, 275 an RDA with the fine-scaled field survey biodiversity data as response variables, habitat 276 heterogeneity, land-use intensity and soils ( influencing the positioning of the sites (scores), which are constrained by the explanatory variables 280 alone ( Figure A2).

281
RDA including all squares in Wales was carried out in R (R core team 2017, Oksanen et al. 282 2017) to enable the use of large datasets. To test the predictive power of the multi-variate analysis, 283 an RDA was carried out with data from the first three years only (2013-2015, 225 1 km squares), 284 using biodiversity metrics from the field survey as response variables and remotely-sensed habitat 285 heterogeneity and land-use intensity data as explanatory variables. Data for all other non-GMEP 286 squares in Wales were passively added to the ordination space using the remotely sensed 287 explanatory variables only. Site scores for year 4 sites (75 squares) based on passively adding them 288 using remotely sensed explanatory variables were extracted. Then the RDA described above was 289 repeated including year 4 field data (2013-2016, 300 squares). Site scores for year 4 squares (75)  290 were extracted from the results of this analysis and compared to the scores extracted from the 291 previous ordination to validate the analysis. 292 Finally, the axis scores from the RDA of all field survey squares (300), with biodiversity 293 response data and remotely sensed explanatory variables and all non-GMEP squares in Wales, 294 passively added to the ordination space, were used to map the extent of HNV land in Wales. 295 296 3. Results 297

Fine-scaled analysis 298
Generalised Additive Models 299 The results of analyses of explanatory variables against species richness can be seen in Table 2 and  300 supplementary Figures A3a-A3e, where the GAM curve has been superimposed onto the raw data. 301 There were no significant relationships with hoverflies. Adding a spatially explicit model to account 302 for SAC did not affect many of the results. Bees were the group most influenced and some results 303 were no longer significant when SAC was accounted for. 304 305 The proportion of semi-natural habitat was positively associated with plant habitat quality 306 indicators, wetland specialist plants, rare invertebrates and upland birds. It was negatively related to 307 butterflies, bees, total plant species richness, woodland butterflies and farmland birds. There were 308 non-linear, unimodal relationships with ancient woodland plants, rare birds and woodland birds 309 ( Figure A3a). There were inverse relationships with the proportion of improved land ( Figure A3b): for 310 example, there were negative relationships for plant habitat indicators, wetland specialist plants, 311 rare invertebrates, and upland birds. 312 313 Habitat diversity ( Figure A3c) was positively, linearly, related to total plant species richness, 314 woodland birds and rare birds and unimodally to bees. There were no significant relationships with 315 the other biodiversity indicators. 316 317 There were positive relationships with broadleaved woodland connectivity ( Figure A3d), for 318 both generalists (butterflies) and specialists (woodland butterflies, birds & plants (slightly u-shaped) 319 and rare birds). Rare invertebrates, upland birds and wetland and plant habitat indicators were 320 negatively related to broadleaved connectivity. Farmland birds and total plant species richness were 321 non-linearly (unimodally) related. There was no significant relationship with bees. The relationships 322 between biodiversity and hedgerow length ( Figure A3e) were quite similar to broadleaved 323 connectivity; the only differences were that total plant species richness, woodland plants and 324 farmland birds were linearly positively related, rather than unimodal, and wetland indicator plants 325 were not significantly related to hedgerows. 326 Multivariate analysis 327 The results of the multi-variate RDA analysis are shown in Figure 1a. Axis 1 and 2 explained 20% and 328 2.7% of the variation, respectively. Axis 3 ( Figure A1) explained 2.3% of the variation. There is a clear 329 gradient between low intensity land-use (high proportion of semi-natural land -HNV type 1) and 330 high intensity land-use (high proportion of improved land) which appears to roughly equate to Axis 331 1, with significant relationships to particular species groups. The other gradient appears to relate to 332 habitat heterogeneity (bottom left to top right) with increasing habitat diversity, broadleaved 333 connectivity, hedgerows and lines of trees on Axis 2. This aligns with HNV type 2. The discrimination 334 of types 1 and 2 HNV was carried out by separately bisecting each of these principal gradients. Since 335 all 1 km squares have a score on each axis, the result is a subset of squares that have the overlapping 336 attributes of both type 1 and type 2 HNV. Thus the two types are not defined to be mutually 337 exclusive when mapped across Wales. 338 In the analysis, the following variables were statistically significant ( (Figure 1a). 346 347 The ordination diagram indicates that (as with the GAMs) a high proportion of semi-natural 348 land was associated with species richness of plant habitat quality indicators, upland birds, wetland 349 plants and rare invertebrates along with wetland connectivity. Broadleaved woodland connectivity 350 was strongly associated with woodland birds, woodland plants, rare birds and total plant species 351 richness. Hedgerow length was positively associated with farmland birds, woodland butterflies, total 352 butterflies and bees. Habitat diversity was strongly associated with woodland birds, rare birds and 353 total plant species richness. 354 355 The association of explanatory variables and response variables enables classification into 356 four quadrants that describe the types of 1 km squares that were found in the data (Figure 1b). The 357 types of square are represented using example 1 km square habitat maps from the field survey. 358 359

National analysis based on remotely-sensed data 360
The results of repeating the ordination of GMEP field data, including explanatory variables from 361 remotely-sensed data, can be seen in Figure A2. There were similar relationships with biodiversity 362 variables regardless of whether they were derived data from field survey or remote sensing. 363 Figure 2 shows the results of testing the prediction of axes scores from a subset of squares 364 using two different methods (with biodiversity data and when passively added to the ordination 365 using only explanatory variables). Figure 2a shows a highly significant relationship between site 366 scores on axis 1 (land-use intensity). The result for axis 2 (Figure 2b) is not significant. 367 368 Axis site scores from the all Wales analysis have been extracted and categorised (based on 369 the 20 th percentile, commonly used to identify upper and lower proportions of distributions whilst 370 not solely identifying the extremes) into 'High' (top 20 percentile), 'medium' (middle 60 percentile) 371 and 'low' (lowest 20 percentile). These have been mapped across Wales (Figure 3), to signify the 372 distribution of Type 1 (% semi-natural vegetation) and Type 2 (habitat heterogeneity) HNV farmland. 373 Figure A4 shows boxplots of the distribution of the ordination axis scores across the categorised HNV 374 classes. The maps suggest that approximately 35% of the land in Wales is in the upper percentile for 375 HNV Types 1 and 2 combined. 376 377 4. Discussion 378 379 Conservation of farmland is important for mitigating biodiversity decline (Kleijn 2011). Identifying 380 areas of High Nature Value spatially enables targeting of conservation actions and farming subsidies 381 (Klimek et al. 2014). In this study, land-use intensity and habitat heterogeneity were clearly 382 identified as two major gradients acting upon species diversity in Wales at the spatial scale of 1 km. 383 They also form the criteria for classification of HNV farmland. Our results therefore provide a 384 uniquely detailed and large-scale test that supports the two hypothesised relationships that define 385 Types 1 and 2 HNV. 386 387

Relationships between land use intensity and biodiversity 388
In Wales, there are large areas of semi-natural, extensively grazed land composed of heathland, 389 semi-natural grassland, bog and purple moor grass rush pasture (Blackstock et al. 2010) and ffridd (a 390 transitional habitat of unimproved grassland, shrub heath, bracken and scrub; Woodhouse et al. 391 2005) and these areas are important in a European context (Russell et al. 2011). They are associated 392 with many habitat-specialist species and are valued for their aesthetic, cultural and functional 393 importance ). This includes upland birds, rare invertebrate species, plants 394 indicative of high conservation value habitats and wetland plants (all of which were significant in this 395 study). It might have been expected that butterflies would be positively related to semi-natural 396 habitat (a number of habitat specialists are only found in such habitats). However, this was not the 397 case. Pollinator surveys were conducted in July and August to coincide with peak butterfly 398 abundance, this is after the main flight period of some Welsh habitat specialists. Also, most habitat 399 specialist butterflies that fly during the survey period have restricted ranges in Wales (e.g. High 400 Brown Fritillary, Argynnis adippe). 401 402 Higher land-use intensity was associated with farmland birds, bees and butterflies, reflecting 403 positive responses of farmland-associated species to a degree of active management. Also, higher 404 land-use intensity tends to be in lowland environments, which have a more benign climate, 405 associated with greater numbers of species. Wales is not as intensively farmed as some countries: 406 there are no large areas of arable, field size is not large and there are often hedgerows and linear 407 habitats, which may explain why species richness among these groups is not lower at higher 408 intensity. In an analysis of all of Great Britain, the relationship between land-use intensity and 409 species richness was unimodal (Maskell et al. 2013). In Wales, land-use intensity is low to medium in 410 comparison to the UK as a whole so it sits on the left and middle of centre of the unimodal curve 411 rather than to the right. 412 413

Relationships between Habitat heterogeneity and Biodiversity 414
Habitat heterogeneity is a desirable cultural landscape quality (Swetnam et al. 2017), regardless of 415 benefits for species diversity. However, both compositional and configurational heterogeneity are 416 positively related to many taxa in landscapes in Wales: habitat specialists (woodland birds, 417 butterflies and plants, rare birds, farmland birds) and generalists (plants, bees and butterflies). There 418 is supporting evidence from the literature: species groups differ in response to environmental 419 heterogeneity (Fahrig et al. 2011). Bees require several different and sometimes also very specific 420 habitat types to persist in a landscape (Billeter et al. 2008). The diversity of butterflies has been 421 shown to be related to small-scale habitat heterogeneity (Weibull et al. 2000). Habitat diversity 422 enables source populations in semi-natural elements to spill over to intensively managed fields 423 (Holland & Fahrig 2000;Smart et al. 2006). Bird species' preferences vary, both with respect to the 424 scale of the heterogeneity and responses to specific levels of heterogeneity (Aue et al. 2014, 425 Siriwardena et al. 2012, Pickett & Siriwardena 2011. There is evidence that bird taxonomic and 426 functional diversity can increase within HNV farmland in relation to land-use composition and 427 increased configurational heterogeneity (Morelli et al. 2018). 428 Woodland varies in extent, condition and distribution across the Welsh landscape (Russell et 429 al. 2011 be spatially precise and sufficiently frequent to detect change (Lomba et al. 2014). It has been 443 possible in some countries to collate the 'best' data available to map HNV farmland as a one-off, but 444 it may not be practicable to repeat this at regular intervals. 445 In this study, we propose using disaggregated fine-scaled data to build models that can 446 include remotely sensed data to provide continuous coverage (Klimek et al. 2014, Boyle et al. 2015. 447 The surveys for data collection can be repeated over set time periods to analyse change. When 448 remotely sensed explanatory variables were jointly analysed alongside field survey data there was a 449 high degree of correlation between them suggesting that there is potential to use remotely sensed 450 data as a surrogate for field survey. 451 Applying this process will be helped by the large volumes of freely available, medium 452 resolution (< 30 m pixel size) satellite data provided by Landsat-8 and Sentinels 1 and 2. These data 453 will lead to more frequent production and updating of EO products, for example the UK Land Cover 454 Map is moving to a three-year repeat cycle, from an approximately 10 year repeat cycle. The 455 increase in the availability of high resolution data from Sentinel-1 and Sentinel-2 is also leading to a 456 wider range of routinely-derived EO-products for the UK, including vegetation productivity (Tebbs et 457 al. 2017). Developments such as these are likely to increase our ability to map HNV and changes in 458 HNV in the future. 459 The testing of the method in the validation analysis demonstrated that the percentage of 460 semi-natural/improved land was a very useful metric for identifying HNV type 1 farmland. However, 461 for habitat heterogeneity and HNV type 2 farmland, although the initial RDA analysis identified some 462 interesting patterns in the species data, the analysis using only remotely sensed explanatory 463 variables to add squares passively did not predict the species richness of the survey squares as well. 464 This may be because multiple explanatory variables were used, rather than one simple indicator, and 465 because of complex relationships between biodiversity and habitat heterogeneity. 466 There are issues with remotely sensed data. Although some habitats can be identified fairly 467 accurately from satellites, e.g. woodland, other habitats (e.g. grasslands, bogs and heath) cannot be 468 classified accurately (Morton et al., 2011, Wood et al. 2018. Vegetation structure can also be 469 difficult to capture remotely: small biotopes (<20 m) which particularly in intensive landscapes may 470 contain valuable biodiversity, will often be below the minimum mappable size of products derived 471 from satellite data (Wood et al. 2018). This may impact on measures such as habitat diversity. 472 Rhodes et al. (2015) found that high-resolution field data generated more reliable models of 473 predicted local population responses to land-use change than lower resolution, remotely sensed 474 data. Further finely scaled analysis at a field level and improvements in remotely sensed data may be 475 necessary to clarify these relationships and to increase explanatory power of the models (Klimek et 476 al. 2014 A high proportion of semi-natural land is associated with high biodiversity of habitat specialists and 480 species indicating areas of high conservation value. This metric can be derived from coarse, remotely 481 sensed data to predict and to map High Nature Value type 1 farmland. 482 483 Habitat heterogeneity is associated with increased diversity of generalist and specialist 484 species groups and interesting relationships were found between broadleaved woodland 485 connectivity, habitat diversity, lengths of hedgerows/lines of trees and field survey biodiversity data. 486 The complexity of these relationships and the inadequacies of current remotely sensed data make it 487 more difficult to replace fine-scaled analysis with simple surrogate metrics. Estimation of extent and 488 spatial configuration of HNV type 2 requires further work to refine the method and to create better 489 metrics. 490 The approach described here, using fine-scaled field survey data collected consistently at 491 frequent intervals in association with remotely sensed data offers a great deal of potential for 492 extrapolating modelled results nationally and also of ensuring repeatability of the analysis to assess 493 change over time, and could usefully be applied to enhance the identification and monitoring of HNV 494 in other European countries. 495 Table 2. Results from GAM's (Poisson distribution) from fine-scaled field data, including spatially explicit model for residual structure; species richness as response variable against explanatory variables. (Dir= direction of relationship, + positive, -negative, ∩ unimodal, U u-shaped. ns= not significant. * p<0.05, **p<0.01, ***p<0.001) 3.9* -12.8*** -ns 18.3*** + 11.4*** -Woodland birds 97.7*** + 17.6*** + 15.9*** + 59.8*** ∩ 28.9*** ∩