Understanding the Performance of Biodiversity Offset Markets: Evidence

1 Understanding the Performance of Biodiversity Offset Markets: Evidence

2 from An Integrated Ecological – Economic Model

4 Katherine Needham (University of Glasgow) 5 Martin Dallimer (University of Leeds) 6 Frans de Vries (University of Stirling) 7 Paul Armsworth (University of Tennessee) 8 Nick Hanley (University of Glasgow)

11 Abstract

12 Biodiversity offset markets can incentivize landowners to take actions that benefit biodiversity. A 13 spatially explicit integrated ecological-economic model is developed and employed for a catchment in 14 the UK where offset buyers (house developers) and sellers (farmers) interact through trading offset 15 credits. We simulate how changes in the ecological metric and geographic scale affects the performance 16 of the offset market. Results show that the choice of the metric has a significant effect on market 17 liquidity and the spatial distribution of gains and losses in the “target” species. The results also 18 consistently reveal relatively higher potential welfare gains for developers than for farmers.

20 We thank The Leverhulme Trust for funding this work under project RPG-2017-148, and members of 21 our advisory group for numerous helpful comments on the research.

27 1 Introduction

28 Biodiversity offsetting is a widespread practice that aims to reconcile the mounting pressure for urban 29 development and agricultural expansion with the need to significantly reduce the impacts of these on 30 biodiversity (Ten, Treweek, and Ekstrom 2010). Biodiversity offsets provide measurable conservation 31 gains to compensate for significant, residual impacts on biodiversity as a result of new development 32 activities (BBOP 2009). The term “offsetting” often encompasses a wide range of mechanisms 33 including compensatory mitigation, in-kind compensation, mitigation banking, habitat banking, species 34 banking, and wetland banking (Lapeyre, Froger, and Hrabanski 2015). As of 2018, 13,000 offset 35 projects have been recorded worldwide covering at least 153,670 km2 (Bull and Strange, 2018). 36 However, the approach is often criticised, since a large number of studies that have shown that offset 37 schemes have failed to deliver no net loss of biodiversity (Robertson 2004; Burgin 2008; Quétier, 38 Regnery, and Levrel 2014). Of particular ecological concern are the implications of offsetting for 39 changes to the functionality of restored systems, the longevity (security) of offset benefits, and time 40 lags between destruction from development and creation of the offset sites (Gibbons and Lindenmayer 41 2007; Maron et al. 2012). Restoration studies have only taken place on a limited range of habitats 42 (grassland and wetlands), which is a key reason why the implications of offsetting for other habitats 43 remain poorly understood (Evans et al. 2014). Perhaps the most contentious area of biodiversity 44 offsetting is what constitutes as an offset, and how to determine ecological equivalence between sites. 45 Across the disciplines of economics and ecology, these issues are seen as critical factors determining 46 the success of offsetting as a policy instrument (Heal 2005; Bull et al. 2013).

47 In this paper we are specifically interested in markets for biodiversity offsets. These are created when 48 multiple buyers and sellers of offset credits interact with others through a trading process. This creates 49 a setting in which landowners can choose to manage land for conservation, generating offset credits 50 which can then be sold to a developer who is required to mitigate development impacts. By establishing 51 an appropriate rate of exchange between sellers and buyers, markets can, in theory, achieve no net loss 52 of biodiversity within some defined area at least cost. By creating economic incentives for conservation, 53 market mechanisms encourage private landowners and firms to take costly actions that benefit 54 biodiversity (Kangas and Ollikainen 2019). Market forces within a biodiversity offset mechanism 55 should steer land allocation choices in the direction of greater efficiency, given some constraint on 56 overall biodiversity change. Buyers will not offer more for a credit than the value to them of land for 57 development, and sellers will require no less than the opportunity costs of creating offsets (through 58 foregoing some other form of land use such as crop production). Such offset markets could therefore 59 allow biodiversity conservation targets to be met at the lowest cost since only those suppliers whose 60 supply prices are competitive will be rewarded with payments for supplying offsets.

61 The objective of this paper is to assess the social welfare implications and ecological impacts of what 62 we consider to be two key design parameters when developing markets for biodiversity offsets: the 63 ecological target which constitutes the biodiversity offset, and the size of the market for trading i.e. the 64 number of players. This is based on the identification of key offset market design parameters in 65 Needham et al. (2019).

66 We first develop a general model of a biodiversity offset market in the context of a no-net-loss 67 regulation. We then construct an integrated ecological-economic model of a biodiversity offset market 68 that builds spatially explicit biodiversity offset supply and demand curves. These curves capture the 69 spatial variations in both the costs of supplying biodiversity offsets and the demand for offsets related 70 to the value of new housing developments across the landscape. Through the identification of the 71 market-clearing price for the biodiversity offset, our model can then be used to assess the economic 72 welfare implications of such a market, and determine who benefits from the design of such schemes. 73 We can also consider the ecological implications of such a market across the landscape to identify where 74 the gains and losses in biodiversity take place. The model is parameterized for a north-eastern region of 75 England: The North Pennines, Tees Valley and North Yorkshire Moors in the context of UK 76 Government’s plan to implement biodiversity offsetting as part of the recently published 25 Year 77 Environment Plan (HM Government 2018). Our model captures (i) how the choice of the ecological 78 metric for trading has significant implications on the scale and distribution of new development, (ii) 79 how the ecological metric has significant impacts on ecological gains and losses of species not covered 80 under the no net loss regulation, (iii) how changing market boundaries changes spatial heterogeneity in 81 the supply and demand curves, and (iv) subsequently shows that a larger geographic scale benefits 82 developers, however, smaller-scale markets are ecologically more beneficial.

83 The remainder of the paper is organised as follows. In the next section, we discuss the related literature 84 and summarise some relevant policy experience with offsetting. In Section 3 we develop our general 85 model of biodiversity offset trading. Section 4 details the empirical approach, including an overview of 86 the integrated economic - ecological model, our case study, data sources and data development. Results 87 are presented in Section 5 and the discussion and conclusion in Sections 6 and 7.

88 2 Biodiversity offsetting: what do we know already?

89 Biodiversity offsetting is the final step in the mitigation hierarchy: this hierarchy is crucial for all 90 development projects aiming to achieve no overall negative impact on biodiversity or a net gain (also 91 referred to as a No Net Loss and the Net Positive Approach). First developers must first actively avoid, 92 minimize and reduce their impacts through restoration actions at the project site (McKenney and 93 Kiesecker 2010). In all cases, biodiversity offsets should meet the criterion of additionality: only those

94 actions that would have not have otherwise occurred should be counted as a biodiversity offset (Laitila, 95 Moilanen, and Pouzols 2014). From a developer’s perspective, it is argued that this market approach is 96 more cost-efficient, quicker and more certain than traditional planning regulations (Levrel, Scemama, 97 and Vaissière 2017, Hannis and Sullivan 2012). The market element allows developers to purchase “off 98 the shelf” offsets where ecological gains have already been established and verified, rather than 99 developing their own restoration sites (Bonds and Pompe 2003). The use of offset banks in the supply 100 of offsets helps deal with one widely-recognised problem with offsets, namely that the ecological 101 consequences or restoration actions are uncertain (Maron et al. 2012).

102 Two key offsets schemes are the US Wetland Mitigation Banking (WMB) market and BioBanking in 103 New South Wales, Australia. The US (WMB) market developed as a response to Section 404 of the 104 Clean Water Act (1977) which required ‘no net loss of wetland acreage and function'. Developments 105 which impacted on wetlands were required to hold credits from investment in completed wetland 106 restoration (Hough and Robertson 2009). The first wetland mitigation banks were established in the 107 1980s. Conservation Banks were approved in the 1990s and sought to protect individual species rather 108 than wetlands functionality. As of 2020, there are currently 1482 mitigation banks and 207 conservation 109 banks in the USA (US Army Corps 2020). WMB credits were initially based on habitat acreage but 110 more recent approaches have sought to take wetland functioning into consideration, accounting for the 111 effects of offsetting actions on water quality, flood storage, and flow reductions and habitat quality. 112 Conservation Bank credits take into account acres of habitat needed to support a particular species or 113 breeding pair.

114 A significant change in the WMB market was the introduction of the 2008 Mitigation Rule which 115 prioritised mitigation banking as the first priority for compensation, ahead of permittee responsible 116 mitigation and out of kind mitigation. This resulted in a significant increase in the purchase of credits 117 and demonstrates how a regulatory change placing emphasis on mitigation banking above other 118 compensatory measures can have a significant positive benefit on market activity. Additionally, a 119 watershed approach to banking was introduced with an emphasis on using existing watershed plans to 120 inform compensatory mitigation decisions. New South Wales, Australia developed BioBanking 121 following the Environment Protection and Biodiversity Conservation Act (1999). The main aim of the 122 scheme is to integrate biodiversity into land-use planning for high growth areas, particularly within the 123 coastal zone (Scanlon 2007). The offset metric is based on the habitat hectares approach, with 124 adjustments to take into account state and national priorities, regional value, landscape value and the 125 threatened species response to proposed management actions (Office of Heritage New South Wales 126 2014). Whilst 53 BioBanking agreements have been secured (as of May 2016) many believe the scheme 127 has not been as successful as expected: the current market has struggled to secure offers from potential

128 suppliers, and thus is failing to produce the cost-savings initially anticipated. In response there are calls 129 to make the scheme mandatory rather than voluntary to encourage offset creation (Dupont 2017).

130 Biodiversity offset markets are in the developmental stage within the UK. Biodiversity is predominantly 131 protected through the European Habitats and Birds Directives (92/43/EEC and 2009/147/EC). However, 132 compensation for damage to sites that are not covered by these Directives is relatively weak (Tucker et 133 al 2013) leading to a decline in biodiversity (Hayhow et al 2019). The UK Government launched six 134 pilot biodiversity offset schemes between 2012-2014. Similar to Biobanking in Australia, the offset 135 metric was based on the habitat hectares approach with trading ratios used to adjust for the risks 136 associated with restoring or expanding certain habitats and to take into account locational differences 137 between the offset site and development site (Defra 2012). Within the UK pilot areas, developers were 138 required to provide compensation for biodiversity loss and could choose to do so through offsetting. 139 Follow-up reviews highlighted that the potential benefits of offsetting would only be realised if driven 140 through mandatory offsetting (Baker et al 2016; Lockhart 2014; Sullivan and Hannis 2015). The 141 approach is currently being revisited as part of the UK Government’s approach to revising land 142 management through the 25 Year Environment Plan (HM Government 2018). This includes embedding 143 the “net gain” principle for development and creating or restoring 500,000 hectares of habitat. 144 Specifically, it identifies that this net gain could be delivered by “habitat banks, land-owners or brokers 145 as part of a flexible market”.

146 Despite the increasing policy interest in biodiversity offsets, and a growing ecological literature, 147 relatively little attention has been paid to the idea by environmental and resource economists when 148 compared to other incentive-based mechanisms, such as payments for ecosystem services and tradable 149 pollution permit markets for air and water pollution (Coralie, Guillaume, and Claude 2015; de Vries 150 and Hanley 2016). From a resource economist's perspective, markets for biodiversity offsets face some 151 of the same design issues as the tradable pollution permit markets including credit definition, market 152 participation, trading ratios and administration (Bonds & Pompe, 2003; Wissel & Wätzold, 2010). 153 Needham et al. (2019) review experience from pollution permit markets and identify some fundamental 154 parameters that seem most relevant for the design of offset markets. The most relevant of these from 155 the perspective of the current paper are:

156 (a) Policy targets and metrics:

157 When establishing a pollution permit-trading scheme, the first stage is for the government or regulatory 158 agency to determine the cap (or limit) on overall emissions. While voluntary trading can occur (for 159 example, in the context of carbon offsets and forest planting), imposing a regulatory cap greatly 160 increases the volume of trading. Efficient environmental markets require goods to be grouped into

161 simple, measurable, standardised units in order so that they are fully exchangeable, i.e., tradable. For 162 biodiversity offsetting, a vague cap dictating “no net loss of biodiversity” leaves room for ambiguity 163 and is difficult to measure and quantify. This undermines effective monitoring and enforcement, with 164 the prospect of poor market functioning in terms of lack of both cost-effectiveness and environmental 165 effectiveness. A clear metric for how offsets are measured is thus required. What exactly this metric 166 should be is much less obvious.

167 (b) The trading ratio:

168 This governs the rate of exchange between offsets at different points in space (between the supply site 169 and the demand site for any prospective trade). Once the metric for an offset market has been established, 170 trading ratios are then required to determine the ecological rate of exchange between sellers and buyers 171 at different points in space. Ideally, such trading ratios need to reflect the many potential sources of 172 ecological heterogeneity implicit in the spatial definition of the credit, and thus the ecological 173 consequences of moving pressures on the biodiversity metric (positive, for offset supply and negative, 174 for offset demand) across different locations within the policy area.

175 (c) Market scale and trading volume:

176 For biodiversity-offset markets, erratic price signals will increase the uncertainty on investor returns 177 from development and, in turn, increase development costs. Higher offset price volatility may also deter 178 potential suppliers from incurring the (opportunity) costs needed to create credits. The implication is 179 that larger offset markets would be preferred to smaller markets on grounds of economic efficiency. 180 However, creating a large market geographically implies a greater heterogeneity in the biodiversity 181 value of conservation actions at different sites, making it harder to measure and ensure equivalence in 182 ecological gains and losses across space. This implies that there will likely be a case-specific optimal 183 trade-off between having a larger spatial scale which increases participation, and a smaller spatial scale 184 which makes the determination of ecological equivalence easier to model and control for via trading 185 ratios. Moreover, ensuring transactions costs of trading are kept as low as possible seems likely to be 186 important.

187 Fernandez and Karp (1998) provided one of the first economic analyses of wetland banking, focussing 188 on the optimal level of investment in wetland mitigation banking in the USA. Using an ecological- 189 economic model they considered how restoration costs, restoration potential and the market value of 190 offset credits determine how much a landowner should invest in a wetlands bank. Following ecological 191 expectations, the authors found that as restoration costs reduce, it is more likely a landowner will restore 192 his whole land parcel. Increases in ecological uncertainty increase the cost of a credit and restoration is 193 harder to justify under higher interest rates. Other economic analyses have focussed on the implications

194 of using different ecological metrics to determine what constitutes an offsets. Heal (2005) compared 195 the habitat hectares approach with species banking to protect the red-cockaded woodpecker and found 196 that more restrictive definitions of the ecological metric increase the price of a credit. Furthermore, 197 choosing the incorrect ecological metric nullifies the ecological benefits of the trading system. These 198 early economic analyses highlight the significant challenges associated with determining the unit of 199 currency when developing an offset market, and how the choice will be likely to significantly affect 200 ecological outcomes.

201 Other modelling approaches have focused on the impact of time lags related to restoration potential of 202 habitats in relation to NNL of species and also their effect on the market (Bendor 2009; Drechsler and 203 Hartig 2011). Time lags have been shown to increase the restoration costs for those wishing to develop 204 biodiversity offsets and greater fluctuations in credit supply prices (Drechsler and Hartig 2011). 205 Solutions to this have been offered regarding the release schedule of credits, with a balance needing to 206 be met between releasing enough “early release credits” (credits which have not yet reached the full 207 restoration potential) to foster investment into offset sites and not over releasing credits which leads to 208 a future net-loss in habitats and species (BenDor, Guo, and Yates 2014).

209 Implications of the trading ratio on market functioning have also been a key theme within the economics 210 literature. Bonds and Pompe (2003) provide one of the first economic analyses for calculating the 211 trading ratio to adjust for the functionality and location of wetland credits. Further work has been 212 undertaken in the ecological literature by BenDor et al (2009) Bruggeman et al (2005) Bull et al (2014) 213 Laitila et al (2014) and Moilanen et al (2009). However, there is still no single accepted method of 214 calculating the trading ratio in biodiversity offset markets.

215 Further research has concentrated on linking economic and ecological models to examine whether offset 216 markets offer opportunities for biodiversity conservation. Doyle and Yates (2010) provide an economic- 217 ecological model of ecosystem service markets following a no net loss principle. A focal point in their 218 analysis is the scale of the market by assessing how (free) entry of mitigators (i.e., suppliers of offset 219 credits) affects ecosystem restoration. Their results indicate that as additional suppliers enter the market, 220 the size of an individual restoration project decreases but the total amount of mitigation increases, thus 221 increasing credit supply. Their results also indicate that using a single metric to capture ecosystem 222 functioning will result in loss of some ecosystem functions, stressing the need to further understand the 223 relationship between restoration and multiple ecosystem service functions. Most recently Kangas and 224 Ollikainen (2019) develop a model to analyse a market for biodiversity offsets with a focus on 225 restoration potential of three habitats: wetlands, forests and agricultural habitats in Finland. Credits are 226 developed as a result of restoration measures which would not have otherwise occurred, and restoration 227 potential is assessed based on expert assessments and the literature. Demand is based on predictions for

228 future land use change. The model considers how market equilibrium adjusts in response to trading 229 ratios, the focus habitat for restoration, and uncertainties associated with restoration outcomes. Results 230 showed that all three of these aspects impacted the market equilibrium price in theoretically anticipated 231 ways: offset prices were higher when restoration actions were more costly and uncertainty high.

232 Building from the above literature, our research uses an integrated ecological-economic modelling 233 approach to explore first theoretically, and then empirically (for a specified case study), the impact of 234 two design parameters on equilibrium for a biodiversity offset market. Based on the identification of 235 key design parameters in Needham et al. (2019), we contrast economic and ecological market outcomes 236 under varying species-specific ecological policy targets and varying market size. Specifically, our 237 modelling approach allows us to develop spatially explicit supply and demand curves for biodiversity 238 offsets. This allows us to test how the choice of the ecological metric alters the market-clearing price 239 of an offset, and subsequently the gains to buyers and sellers from trading. Secondly, we compare the 240 market-clearing price and subsequent welfare implications when the market size is reduced from one 241 large planning region to three smaller regions. This builds on the work of Doyle and Yates (2010) by 242 adding spatial and cost heterogeneity to the model with offset suppliers and developers requiring offsets 243 having varying marginal costs (the empirical modelling by Doyle and Yates assumed homogeneity in 244 costs). In addition, our modelling framework allows us to assess how land-use and specific species 245 distribution and abundance shift within the case study region as a result of offset trading.

246 3 A General Model of Offset Trading

247 Consider a region where land can be divided into three possible uses which are mutually exclusive at 248 any point in time: agriculture, development for new housing, and conservation. Within such a setting, 249 it is assumed that there are a 푛 farmers who own the land and 푚 developers who wish to acquire land 250 for housing development. We further assume that there are a sufficient number of farmers and 251 developers so that no individual in either group has market power, i.e., all are price-takers in their 252 respective output markets as well as the market for offsets.

253 Farmers’ default land use is assumed to be for agricultural purposes in the form of crop or livestock 254 production. However, farmers can offer up some of their land for conservation, where conservation land 255 (newly created wildlife habitat, for instance) earns offset credits depending on the biodiversity metric 256 used. The farmer’s optimization problem entails maximizing the profits derived from agricultural 257 production plus the value of any offset sales. Every hectare given up to create new habitat means one 258 less hectare for agricultural production. We assume that this conversion cost to the farmer consists 259 solely of the opportunity costs of foregone agricultural output. From an agricultural perspective land is 260 of variable quality, measured in terms of agricultural output per hectare. Ranking all land owned by any

261 one farmer along this gradient yields a continuous, upward sloping supply (marginal cost) curve of 262 creating new offset credits. The farmer maximises profits by choosing to supply the quantity of offsets 263 which equates the marginal cost of creating new habitat (i.e., lost agricultural profit at the margin) with 264 the offset price. If agricultural prices rise, then the opportunity cost of creating new habitat rises, and 265 so the supply curve for each individual farmer (hence aggregate supply) would shift up.

266 House developers face land values for housing that varies spatially across the region. Each developer 267 knows that, depending on the conservation value of land, each hectare acquired for new housing 268 development requires a number of offset credits to be purchased from an offset provider or offset bank. 269 Ranking development options according to their expected housing value to this developer from highest 270 to lowest yields a downward sloping demand curve for offset credits. The developer chooses to buy a 271 quantity of credits which equates their individual demand curve with the market price of offsets.1 If the 272 expected average house price in the catchment rises, this would increase the developer’s willingness to 273 pay for any potential housing site, implying a shift of the demand curve for offsets to the right. The 274 slope of the aggregate demand curve depends on the heterogeneity in the derived demand for offsets, 275 i.e., willingness to pay for land for housing development. If each hectare of habitat created can be 276 heterogeneous in ecological terms, then this means that each offset created or demanded has a different 277 ecological value. On the demand side, higher ecological value land requires credits of higher value to 278 be purchased to offset the impacts of development on biodiversity; on the supply side, higher value 279 ecological land generates more valuable credits when new habitat is created.

280 Assume for the moment that we have a single ecological indicator which is the focus for policy. Then

281 we can think of the value of any given credit supplied to the market to be ℎ푒푖, with ℎ referring to the

282 amount of hectares and 푒푖 indicating the ecological value at site 푖. On the demand side, a developer

283 must obtain a number of offsets equal to a value ℎ푒푗, where 푒푗 reflects the ecological value of site 푗 284 which is damaged by the development. Assuming a development of one hectare in size, the developer

285 must ensure it purchases appropriate offsets such that ℎ푒푖 = ℎ푒푗. From this we can define the offset 286 trading ratio between any two sites 푖 (supply side) and 푗 (demand side) as:2

푒 287 (1) 푒 ≡ 푖. 푒푗

1 This could mean that some developers choose to buy no credits and not develop any housing. 2 If there is more than one ecological indicator, then there are multiple trading ratios between sites, one for each target.

288 We are now in a position to develop a simple general market equilibrium framework to highlight the 289 core functioning of the market for biodiversity offsets. This not only allows us to draw some clear 290 comparative statics results, but also provides a solid basis for the actual case study implementation later. 291 We will do so by building on Doyle and Yates (2010) but extend their model in two important ways. 292 First, our model captures heterogeneous opportunity cost of land use. Second, given this heterogeneity, 293 we will derive explicit demand and supply functions for three wading bird species.

294 Given a number of 푛 farmers, the quantity of hectares (offsets) “produced” by an individual seller can

295 be represented by the function ℎ(푛, 푒, 푝, 푐푘), where 푝 is the offset price, 푐푘 is the opportunity cost of 296 land use of type 푘. Aggregating across all farmers, the total amount of land supplied for conservation,

297 퐻푆, is then:

298 (2) 퐻푆(푛, 푒, 푝, 푐푘) = 푛ℎ(푛, 푒, 푝, 푐푘).

299 On the demand side there are a total of 푚 developers. These developers must buy a number of hectares 300 (offsets) to offset the degradation of each hectare ℎ due to development. More specifically, for each 301 hectare degraded it must buy an amount 푒ℎ, where ℎ(푚, 푒, 푝, 푟), where 푟 is the value of land for housing 302 in the development area and 푒 the trading ratio as specified in Eq. (1). The total amount of hectares

303 demanded, 퐻퐷, for development is then:

304 (3) 퐻퐷(푚, 푒, 푝, 푟) = 푚푒ℎ(푚, 푒, 푝, 푟).

305 The aggregate demand curve for offset credits, representing a particular set of spatial values for the 306 value of land for housing development, and the aggregate supply curve for offsets, showing the value 307 of foregone profits from farming, the offset market clears at an equilibrium offset price ∗ ∗ ∗ 308 푃 (퐻 (푛, 푚, 푒, 푐푘, 푟)). This generates an equilibrium amount of hectares 퐻 for biodiversity offsets

309 such that 퐻푆(푛, 푒, 푝, 푐푘) = 퐻퐷(푚, 푒, 푝, 푟).

310 Knowing the expression of the equilibrium price of offset credits, 푃∗, we can do some comparative 311 statics exercises to determine how the equilibrium price responds to changes in the identified 312 determinants. First, a change in the market size governs the offset equilibrium price in a straightforward 313 manner. An increase in the number of sellers, 푛, would increase the amount of hectares supplied to the 314 market for offsetting, which would have a downward pressure on the equilibrium price. In contrast, the 315 equilibrium price would increase if demand would be enhanced through an increase in the number of 316 buyers, 푚. These two effects are summarized by Eqs. (4a) and (4b), respectively:

휕푝∗ 휕푝∗ 휕퐻∗ 317 (4푎) = < 0 휕푛 휕퐻∗ 휕푛

휕푝∗ 휕푝∗ 휕퐻∗ 318 (4푏) = > 0 휕푚 휕퐻∗ 휕푚

319 Second, a rise in the opportunity cost of land use of type 푘, implying more profit foregone from 320 agricultural production, has a suppressing impact on the total amount of land supplied at any price. This, 321 in turn, would push up the equilibrium price of offsets for a given demand for offsets: 322 휕푝∗ 휕푝∗ 휕퐻∗ 323 (4푐) = ∗ > 0 휕푐푘 휕퐻 휕푐푘

324 Third, higher expected house prices would attract more land being demanded for development. For a 325 given supply of land, this would increase the equilibrium offset price:

휕푝∗ 휕푝∗ 휕퐻∗ 326 (4푑) = > 0 휕푟 휕퐻∗ 휕푟

327 Finally, a change in the trading ratio between site 푖 and 푗 can originate from a change in the ecological 328 value of either site. More specifically, following the definition of the trading ratio in Eq. (1), an 329 increasing trading ratio can be the result from a higher (lower) ecological value of site 푖 (푗). If the 330 ecological value of site 푖 increases, this reduces the number of offset credits a developer needs to buy, 331 since each credit now delivers more biodiversity.. The same effect applies if the ecological value of site 332 푗 decreases. In this case too the developer needs to buy less credits to offset each hectare of land on 333 which development occurs, since the land converted to development is ecologically less valuable. This 334 reduction in offset credits demanded implies that the equilibrium offset price decreases with an 335 increasing trading ratio:3

휕푝∗ 휕푝∗ 휕퐻∗ 336 (4푒) = < 0 휕푒 휕퐻∗ 휕푒

337 4 An Integrated Ecological – Economic Model of a Biodiversity Offset Market

338 4.1 Model Overview

339 In the previous section, we developed a general model for a biodiversity offset market. We now adapt 340 this model and apply it to our case study, using integrated ecological-economic modelling. Our

3 In a similar vein, the opposite effect occurs when the trading ratio as defined in Eq. (1) decreases through either a decrease (increase) in the ecological value of site 푖 (푗). In both these situations the developer’s demand for offset credits increases, pushing up the equilibrium offset price.

341 integrated model seeks to maximise the joint agricultural and development profits of landowners across 342 a case study region, subject to a regulatory limit of no net loss of the ecological target. Each landowner 343 manages a single 1 by 1 km land parcel and has three options: develop land for housing, provide 344 biodiversity offsets, or maintain the current land use. We assume that landowners maximise profits from 345 land use for each parcel. This profit maximization is subject to a number of constraints: land that is 346 already developed for housing can only remain in a developed state; agricultural land can either remain 347 as agricultural land or the farmer can undertake actions which benefit the ecological target and thus 348 generate biodiversity offset credits. If a landowner chooses to develop, he must hold the relevant number 349 of offset credits to ensure no net loss of the ecological target in the land parcel.

350 Our empirical approach consists of three stages. Firstly, an ecological model predicts the current level 351 of our ecological target variable across the case study region, based on current land use. This provides 352 us with a no net loss baseline for the ecological target. Secondly, the ecological model is used to estimate 353 potential changes in the ecological target as a result of landowners undertaking actions which benefit 354 the ecological target, thus generating the potential number of biodiversity offset credits a land parcel 355 could supply. We then determine the profitability of each land parcel under each land-use options, i.e., 356 development, offset provision or current land use. By integrating this profitability with the offset 357 requirements, supply and demand for offset credits for each land parcel is determined. Finally, we model 358 optimal, sequential trades (through a Walrasian auctioneer) based on derived spatially derived demand 359 and supply curves and the no net loss policy goal. This establishes a market-clearing price for the 360 biodiversity offset. A schematic of the modelling process is provided in Figure A in Appendix 1.

361 4.2 The Case Study

362 We apply our model of biodiversity offsetting to a UK case study region broadly known as the Tees 363 Valley, Pennine Uplands and North York Moors (Figure 1) The case study region covers an area of 364 approximately 5400 km2 and encompasses a range of habitats and land use types, from upland moors 365 in the west of the region, low lying agricultural land throughout the central region and increasing 366 urbanization in the east at the coastal margin. The case study region includes three Special Protection 367 Areas (SPAs) designated under the EU Birds Directive and three Special Areas for Conservation (SACs) 368 designated under the EU Habitats Directive. The North Pennine Moors SPA and SAC covers 1030 km2 369 and is designated for its upland dry heaths (heather and blanket bogs) and calcareous grasslands. The 370 North Yorkshire SPA covers 441 km2 and is designated for its populations of Merlin (Falco 371 columbarius) and Golden Plover (Pluvialis apricaria). In addition, it contains the largest tract of heather 372 moorland in England. The North Pennine Moors SPA encompasses extensive tracts of semi-natural 373 moorland habitats including upland heath and blanket bog. It is designated for its populations of Hen 374 Harrier (Circus cyaneus), Merlin (Falco columbarius), Peregrine (Falco peregrinus) and Golden Plover

375 (Pluvialis apricaria). The Teesmouth and Cleaveland Coast SPA is classified for its breeding Little tern, 376 passage Sandwich tern, wintering Knot and Redshank and an assemblage of over 20,000 wintering 377 waders (JNCC 2020).

378 The case study was chosen based on the development pressures within the Tees Estuary portion of the

379 case study which includes the Teesmouth and Cleveland Coast SPA (Planning area 1 in Figure 1). The 380 Stockton-On-Tees Economic Strategy (2017 – 2032) identifies the need to maximise the river and port 381 as key economic assets (Stockton-On-Tees Borough Council 2016) with further expansion of the port 382 infrastructure and the chemical and process industries which it supports, placing further pressures on 383 these coastal habitats (Simpson 2011). Within the Tees Valley, an increase in 27,000 households is 384 projected by 2039 requiring an increase in the affordable housing stock of 1,832 to 2,106 dwellings per 385 annum (Ferrari and Dalgleish 2019). Amongst this intensive development pressure is the SPA made up 386 of a complex of discrete sites, with additional non-designated areas also used for foraging and roosting 387 by wading birds. The area has been highly modified by human activities, with over 90% of intertidal 388 habitats lost to land claim (Smith 2011). The competing pressures for development and conservation of 389 biodiversity make this an ideal study region in which to test our general model of biodiversity offsets.

390

391 392 Figure 1: Tees Valley, Pennine Uplands and North York Moors case study region. Inset map depicts the 393 location of the case study on the North East coast of England. A landowner in our study correlates with a 394 single 1 by 1 km grid or land parcel. 13 Local authorities are labelled on the map which forms the three 395 planning regions: Planning area 1 (Hartlepool, Darlington, Stockton on Tees, Middlesbrough and Redcar

396 and Cleveland), Planning area 2 (Northumberland, Sunderland, Country Durham and Eden) and Planning 397 area 3 (Richmondshire, Hambleton, Ryedale and Scarborough). Contains Natural England data on English 398 SPAs issued under Open Government licence v3.0 unless otherwise stated © Crown Copyright 2019. 399 Contains OS data 1:250 000 Scale Colour Raster © Crown Copyright 2019.

400

401 4.3 Market Design Parameters

402 As highlighted in the introduction, the first stage of developing a biodiversity offset market should be 403 the identification of the policy target in order to determine the ecological metric which denominates an 404 offset credit. Based on the international importance of the case study region for supporting breeding, 405 roosting and overwintering bird populations, the policy target for our modelling approach is NNL of 406 each of three wading bird species: the Eurasian oystercatcher (Haematopus ostralegus), Eurasian 407 curlew (Numenius arquata) and the northern lapwing (Vanellus vanellus). These species were chosen 408 based on national priorities for wader conservation: all three species are recorded as ‘Near Threatened’ 409 on the international IUCN Red List4 of Threatened Species (Bird Life International 2019) Of particular 410 concern is the decline in curlew populations, with UK populations declining at the steepest rate recorded 411 throughout the curlew’s range in Europe. Population declines are linked to a loss and degradation of 412 breeding habitat as a result of changing agricultural practices, such as the conversion of traditional 413 meadows to intensive grassland monocultures (Franks et al. 2017). One model of biodiversity offsetting 414 will be constructed for each species separately, that is each model focusses on no net loss of a single 415 target species. This follows the recommendations of Needham et al (2019) which require that the 416 biodiversity offset market to be based on a simple, measurable ecological metric which can be 417 denominated to standardised units. This approach allows us to ensure no net loss of the target species, 418 but to also consider what we call the “unintended consequences of trading”: what are the ecological 419 impacts on the two species not considered under the no net loss policy target?

420 In addition to our policy target, the regulator also sets a “preferred land management practice” for 421 generating offset credits. The land management practice is one which is expected to benefit the target 422 species. By replacing the current land use with this new, ecologically-preferred land management 423 practice, the expected abundance of the target species will increase, hence generating biodiversity offset 424 credits. The number of credits created in each parcel of land depends on the “ecological productivity” 425 of that parcel, defined as the estimated increase in abundance of the target species if a parcel at a specific 426 location is converted to this preferential land use type. For our model, we adopt a preferred land 427 management practice of “improved grassland” without livestock. This change in land management 428 practice will only take place on agricultural land patches currently farmed for crops or livestock, as such 429 improved grassland is more beneficial to the three target species than the agricultural practices of crop 430 and/or livestock production.

431 We compare the effects of market size on the ecological and economic efficiency of the offset market 432 by conducting our analysis at two spatial scales for the oystercatcher policy target. The analysis is first 433 undertaken across the full case study region so that only one market exists. Three further models are 434 then run using “planning areas” as sub-markets.

435 We do not incorporate temporal dynamics within our model, instead, we compare the case study region 436 in a state prior to offsetting and a state where full offsetting has taken place up to the market equilibrium. 437 We capture spatial heterogeneity in the definition of the credit by integrating a predictive ecological 438 model with our economic model of the offset market. A detailed discussion on the ecological modelling 439 is provided in the next Section.

440 4.4 Data Development

441 The spatial resolution used in the analysis are land parcels of 1 x 1 km aligned to the Ordnance Survey 442 British National Grid, with each 1 x 1km parcel representing a single landowner. The case study region 443 contains 5280 parcels and each land parcel contains data from four spatially referenced datasets 444 covering land classification, crop distribution, housing values, and wading bird abundance and 445 distribution. Where possible datasets from 2017 are utilised. Detailed information on data development 446 can be found in Appendix 1 of the paper. A summary is provided in the following paragraph.

447 For each sampled 1 x 1 km parcel, ESRI ArcGIS v10.4 (ESRI, Redlands, CA) was used to derive 448 information on the UK Biodiversity Action Plan Broad Habitats classes and crop cover. Land is 449 classified using the Centre of Ecology and Hydrology (CEH) Land Cover Map 2015 (LCM2015) and 450 CEH Land Cover plus 2017 (Rowland et al 2015). This produces 21 land classifications following the 451 UK Biodiversity Action Plan Broad Habitats classes including urban, improved grassland, arable and 452 horticulture. Arable and horticulture is further sub-divided into 11 crop types (beet, field beans, 453 grassland, maize, oilseed rape, potatoes, spring barley, winter barley, spring wheat, winter wheat 454 (including oats) and other). The value of agricultural land was calculated using gross margin data for 455 crops and livestock from the SRUC Farm Management Handbook (Beattie 2019). For crop data, the 456 crop coverage derived from the Land Cover plus Crops dataset was combined with the crop gross 457 margin data from the SRUC Farm Management Handbook. Gross margin values are given for each crop 458 type per ha on a productivity range from low, medium to high yield. To account for differences in yield, 459 data on soil quality at the 1 by 1 km resolution was derived and linked to the crop distribution. For 460 livestock, data was sourced from the England Agricultural Census which details the number of livestock 461 by type at the 5 by 5 km grid resolution. Using the livestock gross margin data from the SRUC Farm 462 Management Handbook, the total gross margin for all livestock at the 5 by 5 km resolution was 463 calculated. This was then disaggregated to the 1 by 1 km resolution.

464 Data on actual housing land values in the UK is limited due to the highly competitive nature of land 465 sales. Instead, we specify the value of land for housing development using the UK Government’s House 466 Price Paid dataset as a proxy. HM Land Registry publishes transaction data for all residential properties 467 sold from 1995 onwards and this is made available at the most detailed postcode level for England. 468 Property prices were first converted in 2017 prices using the Bank of England’s RPI figures (this is the 469 preferred inflation measure for house prices within the UK as used in the House Price Index4). The 470 average house sold price for each postcode sector was then calculated. This data was overlaid with the 471 1 x 1 km land parcels in ArcGIS. This allowed the average sold price for each land parcel to be derived. 472 From government and industry reports, it has been estimated that land value represents between 30% 473 and 50% of the price paid for residential property (HM Government 2018). Consequently, in our 474 modelling, we take the land value at 50% of the mean house price in each gird square.

475 Species data is obtained from the British Trust of Ornithology (BTO) Breeding Bird Survey (BBS). 476 Data was received for the Eastern region of the UK for the sampling years 2016 and 2017. There is no 477 dataset available which details abundance of the target species for every 1 by 1 km land parcel within 478 the case study region. Consequently, as part of the ecological modelling process, we developed a 479 Species Abundance Model (SAM) for the eastern region of the UK which could then be used to predict 480 the current distribution and abundance of the target species for every grid square in the case study region 481 based on current land use.

482 SAMs generate predictions of abundance for unsampled locations in a study area using existing data 483 from other sample areas, extrapolating this to new areas based on environmental characteristics (Barker 484 et al. 2014). The development of these ecological models allowed us to obtain predictions for the 485 abundance of the three target species based on current land use and any land-use change as a result of 486 biodiversity offsetting. The first stage of the modelling process was to produce a multiple regression 487 model to identify which parameters (environmental characteristics) predict current distributions for 488 each of the target species (curlew, oystercatcher, lapwing) across the eastern region of the UK. This 489 allowed us to identify which land management practices are most beneficial to the target species (related 490 to the parameters used in the modelling process) and thus serve as our “preferred” land management 491 practice within the biodiversity offset market. The SAM could then be used to predict how the

4 The index applies a statistical method, called a hedonic regression model, to the various sources of data on property price and attributes to produce estimates of the change in house prices each period. The index is published monthly. More detail available from: https://landregistry.data.gov.uk/app/ukhpi.

492 distribution and abundance of the target species could alter as a result of land-use change at a 1km grid 493 square level, allowing us to generate the predicted biodiversity offset credits.

494 The choice of a specific functional form for SAMS is governed by the knowledge of the species and 495 characteristics of the available data. The majority of SAMS sit within the generalised linear model 496 framework (GLM). Such models are typically described in terms of their systematic component in 497 which the response is linked to the environmental data, and their stochastic structure that describes the 498 error distribution (Venables and Springer 2002). For our analysis, we employed the negative binomial 499 model. This is a count data model, derived from the Poisson model but accounts for over dispersion

500 within the data (Potts and Elith 2006). Let us denote as a random variable the single species count, Yij, 501 from the ith site and the jth count. We assume:

502 (5) 퐸(푌푖푗) = 휇푖푗.

503 We allow the mean to be a positive function of covariates

′ 504 (6) 휇푖푗 = exp(풙푖푗휷),

505 where 풙푖푗 is a vector of measured covariates for the jth count of the ith site, and 휷 a vector of parameters. 506 We fit the following model for each regression method:

′ 507 (7) 푥푖푗훽 = 훽0 + 훽1푥푖1 + 훽2푥푖2 + ⋯ + 훽푛푥푖푛 + 휀푡.

508 The models were fit by minimizing the negative log likelihood using the minimization procedure (glm2) 509 in the software package R (Version 3.6.2). The best model for each species was selected by comparing 510 the AIC values (Burnham and Anderson 1998). Models with the smallest AIC value are considered the 511 best model. The model was also evaluated for a sub-sample of the 2016 data set which was independent 512 of the dataset used the primary modelling. Explanatory variables for the ecological modelling were 513 derived using a series of spatially referenced data products: habitats were described using the land 514 classification data derived from the LCM2015 and the Land Cover plus Crops map. A further set of 515 environmental variables were derived using OS Open Map products. These variables were chosen based 516 on the existing literature for predicting bird habitat suitability (Brotons et al. 2004; Tattoni, Rizzolli, 517 and Pedrini 2012).

518 The results highlight that none of the crop types are positively associated with increased abundance of 519 any of the focus species, however switching from the crop types most negatively associated with each 520 species to improved grassland could benefit each species, dependent on other environmental factors 521 within the 1 by 1 km land parcel (the ecological modelling results can be found in Table A in Appendix 522 2).

523 Once the preferred predictive model was established for the eastern region data set, predictions could 524 then be made for the target species abundance for all land parcels across the case study region based on 525 current land use and changes in land use (Table 1). This allows us to identify our offset losses and gains. 526 For each target species, the predictive model calculates the current species abundance and determines 527 how this abundance would change based on all landowners in a region adopting the preferred land 528 management practice for the target species. The chosen land management practice is switching any land 529 patch containing crops to improved grassland.

530 Table 1: Predicted species abundance within each 1 x 1km land parcel across the case study region based 531 on current land use and under a full offset scenario (the full offset scenario is based on all crop hectares 532 being converted to improved grassland)

Predicted species abundance within each 1 x 1km land parcel based on current land use Median count of birds Total predicted abundance Total number of in a single land parcel Bird species across case study region (95% parcels with presence with presence (min – confidence intervals) max) Lapwing 9267 (8962 - 9572) 4764 1.6 (0.1 - 62.0) Oystercatcher 4301 (3865 - 4737) 4860 0.4 (0.1- 45.0) Curlew 8986 (8697 - 9276) 4252 1.5 (0.1 - 40.3) Predicted species abundance under full offset scenario (based on all crop ha being converted to improved grassland) Median count of birds Total predicted abundance Total number of in a single land parcel Bird species across case study region (95% parcels with presence with presence (min – confidence intervals) max) Lapwing 12485 (12165 - 12803) 4806 2.3 (0.1 - 65.4) Oystercatcher 5834 (5324 - 6344) 5138 0.8 (0.1 - 74.3) Curlew 11149 (10883 - 11416) 4762 2.0 (0.1 - 40.4)

533 4.5 The Empirical Modelling Framework for the Offset Market

534 As a baseline, we take the current land use structure in the case study area of size 퐿. This is divided into 535 a number of 푛 land parcels equalling a size of 100 ha per parcel, i.e., a parcel is measured at a resolution 536 of 1 × 1 km. Each land parcel 푖 = 1, … , 푛 can comprise of any combination of 30 distinguished land-

537 use types and crop classifications, denoted 퐴1, … , 퐴30. A policy target is chosen based on no net loss in 538 abundance of a single wading bird species known as our ecological metric (the baseline being set by

539 the current population of this species). Each land parcel has an associated ecological index 푒푖, which is 540 the abundance of the single species found within the corresponding land parcel. Across the whole case 푛 541 study area, total species abundance is then simply the aggregate 퐸 = ∑푖=1 푒푖, which represents the no 542 net loss conservation objective.

543 The regulator sets the preferred land management practice that benefits the target species and this allows 544 us to identify the land parcels that offer biodiversity offsets. Offsets are generated based on a farmer

545 switching the entire land parcel to the new land management practice. This provides us with a measure 546 of single species abundance for each land parcel before and after the change in land management 547 practice. This subsequently allows us to calculate the gains and losses for the species within the land 548 parcel. Recall that we assume that the default crop and livestock distribution on agricultural land is 549 currently the most profitable to a farmer, and that switching to an alternative land management practice 550 will (potentially) result in a loss of profit. Consequently, for a farmer to choose to supply offsets, the 551 offset price must at a minimum compensate for this loss in profit. We can then calculate the farmer’s 552 minimum willingness to accept for a change in land management practice, which gives us the minimum 553 unit price farmer 푖 = 1, … , 푛 would be willing to supply one offset:

푚푖푛 푃 554 (8) 푊푇퐴푖 = 푃푖 = . 퐵푖

푚푖푛 555 With 퐵푖 denoting the increase in the target species abundance gained from offsetting, 푃푖 thus reflects 556 the unit price of a single wading bird.

557 This allows us to generate a series of supply prices for a single offset for each land parcel. To do so, we 558 first calculate the farmers’ profit by using crop and livestock gross margins (revenues minus variable 559 costs) data within each land parcel 푖 = 1, … , 푛:

퐾 푡 푡 푡 560 (9) Π푖 = ∑ 퐴푘휋푘 푘=1

푡 561 where 휋푘 is the gross margin for crop and livestock type 푘 = 1, … , 퐾 in period 푡 = {0,1}, with 푡 = 0 562 and 푡 = 1 referring to the corresponding values before and after changing the agricultural land 563 management practice, respectively. The difference in the gross margin reflects the opportunity cost and 564 is the compensation required by the farmer in order to enter the market as an offset supplier:5

0 1 565 (10) 푃 = ΔΠ푖 ≡ Π푖 − Π푖 .

566 Knowing the opportunity cost that follows from (10) and taking the predicted increase in the abundance

567 of the target species 퐵푖 from the ecological model, land parcels can now be ranked based on which 568 deliver the “best” value biodiversity offsets. Substituting these values into (8) gives a straightforward

5 Note that whilst the farmer will lose the gross margin for the original crop (the opportunity cost), still a profit on the replacement crop will be made after switching.

569 basis on which to rank of parcels, where the inverse of (8) is essentially a benefit/cost ratio of gains in 570 the target species.

571 On the demand side of the market, developers’ demand for offset credits will be determined by the 572 expected revenue from converting their land into housing, taking into account the need to purchase 573 credits to offset any losses in the target species:

푟 574 (11) 푃푚푎푥 = 푗 푒ℎ

푚푎푥 575 with 푃푗 referring to the maximum willingness to pay for an offset by developer 푗 = 1, … , 푚.

576 For each land parcel we have information on the average house price sold and the proportion of the 577 parcel currently converted to housing. For consistency with the agricultural gross margin data, we 578 transform the house price sold data into a rental value of the land for the developers. To derive this land 579 rental value we first determine the remaining hectares of land within a parcel which can be converted 580 to housing development. Taking the 100 ha as the base of the parcel level and adjusting for the total 581 amount of land devoted to urban (퐻푢푟푏) and suburban (퐻푠푢푏) land use, the remaining land available for 582 development is:

583 (12) 퐻푑푒푣 = 100 − 퐻푢푟푏 − 퐻푠푢푏.

584 Dividing the average house price sold on a land parcel by 100 gives us the average house price sold per 585 hectare, 푟̅ℎ푎 . Using this value and (12) allows us to determine expected house price for the yet 586 undeveloped hectares within a parcel, i.e., the expected value of land sold for potential housing 587 development:

588 (13) 피[푟] = 훼푟̅ℎ푎퐻푑푒푣,

589 where 훼 ∈ [0,1] is the fraction of a house price that accounts for the value of land.

590 As a final step, we derive the expected net development value of land for housing, 푉, by adjusting the 591 land rental value obtained in (13) for the gross margin from the agricultural production within the land 592 parcel:

퐾 1 593 (14) 피[푉] = 피[푟] − ∑ 휋푘 푘=1

594 We assume that a developer will convert all remaining hectares of a land parcel to housing, subsequently 595 losing any income from agriculture and requiring the purchase of biodiversity offsets. For development

596 to take place, the developer must hold biodiversity offsets equal to the predicted abundance the target 597 species which are lost as a result of converting his land parcel to housing. As a result, the maximum 598 willingness to pay for a single offset for each land parcel is equal to the net value of land for housing 599 (14) relative to the amount of offsets required:

피[푉] 600 (15) 푊푇푃 = 푃푚푎푥 = . 푗 푗 푒ℎ

601 We now have a range of offset prices based on farmers’ minimum WTA (8) to change the land 602 management practice and the developers’ maximum WTP (15) for a single offset, from which we can 603 generate the supply and demand curves of offset credits. For each price point, farmer 푖 = 1, … 푛 will

604 choose to supply offsets ℎ푖 if the supply price is at least equal to the unit cost of providing an offset,

605 i.e., its 푊푇퐴푖. This is summarized in the following general decision rule guiding a farmer’s decision:

푚푖푛 퐵푖 if 푃 ≥ 푊푇퐴푖 = 푃푖 606 (16) ℎ푖 = { . 0 otherwise 607 608 Similarly, for each price point developers will choose whether or not to purchase offsets, and if so, how 609 many. In this respect, note that a developer must develop a whole parcel and must acquire a sufficient

610 quantity of offset credits ℎ푗 to replace all lost species within the parcel. More specifically, the demand 611 for offsets by developer 푗 = 1, … 푚 is governed by the following general decision rule:

0 푚푎푥 푒 if 푃 ≤ 푊푇푃푗 = 푃푗 612 (17) ℎ푗 = { . 0 otherwise

613 We can now determine the quantity of offsets that are demanded and supplied at each price point across 614 all land parcels, yielding the offset supply and demand curves and allowing the identification of the 615 offset equilibrium price. Once the market equilibrium is established, straightforward welfare 616 assessments can be done by computing standard producer (farmer) surplus and consumer (developer) 617 surplus for the various scenarios tested.

618 4.6 Model Assumptions

619 Our empirical framework makes a series of assumptions. Firstly, our model is run over two time periods 620 t={0,1}, with t=0 and t=1 referring to the corresponding values before and after changing the 621 agricultural land management practice. Consequently, neither the ecological or economic models take 622 into account temporal dynamics. This implies that there are no time lags between the time period where 623 we lose abundance as a result of development and the time period where we gain abundance as a result

624 of the land management practice change. In practice, we would expect the losses to take place relatively 625 quickly once the development of a site begins, but the increases in abundance to happen gradually over 626 multiple months and years. We also recognise that there could be time lags between a species moving 627 between the site impacted by development and the newly created offset site. Secondly, we assume that 628 the current land management practice for farmers (i.e., current crop rotation or livestock) is privately 629 optimal and that switching to a new land management practice will result in a net loss in agricultural 630 profits. We also assume that once land has been secured for biodiversity offsetting it cannot be changed 631 into different land uses and it is protected in perpetuity. Finally, we assume not all land use types can 632 be developed new housing. The non-developable land types are coastal habitats (saltmarsh, littoral rock 633 and sediment, supralittoral rock and sediment), inland rock and woodland (coniferous and broadleaf). 634 We address the implications of these assumptions in the discussion.

635 5 Results

636 5.1 Outcomes of Offset Trading

637 The first design parameter we are interested in is how the choice of the biodiversity policy target, and 638 thus the unit of exchange, effects the market-clearing price, economic surplus and the ecological 639 landscape. The market-clearing equilibrium price of a credit varies between the three different policy 640 targets (Figure 1). The costliest offset credit is for a single oystercatcher credit under the NNL of 641 oystercatcher policy target. An oystercatcher offset credit costs £21,860 per bird, followed by lapwing 642 (£12,600 per bird under NNL of lapwings as the policy target) and curlew (£9971 per bird under NNL 643 of curlew as the policy target). We find that the greatest amount of housing development took place 644 under the oystercatcher policy target with 694 land parcels developed. Of these, 661 land parcels 645 required the developer to purchase offset credits. In contrast, under the lapwing target, 161 land parcels 646 were developed, and 81 required lapwing offset credits to be purchased. Under the curlew target, the 647 smallest amount of development took place, with 108 land parcels developed, of which 61 required 648 curlew offset credits to be purchased. From the supply perspective, there was the highest number of 649 suppliers for the oystercatcher target with 27 land parcels changing their land management practice to 650 become offset suppliers. 24 land parcels changed their land management practice to supply lapwing 651 offsets (under the lapwing policy target) and six land parcels changed their management practice to 652 supply curlew offsets (under the curlew policy target).

653 654 Figure 2: A comparison of the market-clearing price, the number of land parcels developed requiring 655 offsets and the number of land parcels choosing to supply offsets for each policy target

656

657

658 We can also consider view these results spatially (Figure 3). Figure 3 highlights the locations of the 659 newly created offset supply sites and where we see declines in the policy target as a result of 660 development taking place within the land parcel. At the market equilibrium, there was NNL in 661 abundance of the policy target for the three different species. For the curlew policy target, six land 662 parcels changed land management practice to become offset supply sites. These land parcels are located 663 at the coastal margin and each parcel provides between one and four offsets (which can be thought of 664 as increase between one and four curlews as a result of the change in land management practice). In 665 contrast, new development takes places further inland away from the coastal margin, with the greatest 666 number of curlews lost to development in a land parcel being two. For the lapwing policy target, the 667 maximum number of lapwing offsets supplied by a single land parcel is six. The offset supply sites for 668 lapwings are geographically closer to where the development takes place compared to the offset supply 669 sites for curlew (under the curlew policy target): the majority of the lapwing offset supply sites are 670 within 5 km of where lapwings are lost as a result of land parcels becoming developed. In contrast, 671 curlew supply sites are found at least 15 km from where curlews are lost as a result of new development. 672 Oystercatcher supply sites hold the greatest concentration of offsets, with some of the offset supply sites 673 providing more than 10 offsets per land parcel (which can be thought of as an increase of more than 10 674 oystercatchers due to the change in land management practice). The oystercatcher supply sites are 675 geographically distant from where oystercatchers are lost as a result of new development, with the 676 greatest distance between offset supply site and development impact over 100 km.

677

678 Figure 3: A comparison of biodiversity offset supply sites and development impacts under the three policy 679 targets: curlew (top left panel), lapwing (top right panel) and oystercatcher (bottom left panel).

680 Across all policy targets, there are gains from trade realised as a result of trading biodiversity offsets 681 compared to a no-offset-trading scenario. The greatest gains from trade are realised under the 682 oystercatcher policy target. Across all policy targets, consumer surplus accruing to developers is 683 consistently greater than producer surplus earnt by offset suppliers, suggesting the offset trading offers 684 more benefits to the housing developers than agricultural landowners. Despite these potential gains 685 from trade under the offset market scenario, a small proportion of potential development land was 686 utilised for new developments. The highest number of new developments took place under the 687 oystercatcher metric, with 15% of all possible development area having new housing built, compared 688 to 2% of the land area under the lapwing metric and less than 1% of land area under the curlew metric.

689 We can compare the impacts of trading on the non-target species. By this we mean that if the policy 690 target is NNL of oystercatchers, do we see a positive or negative change in abundance of lapwing and 691 / or curlew (the non-target species in this scenario) as a result of offset trading? The results show, that 692 whilst the biodiversity offset market secures NNL of the target species under each of the single species 693 targets, there is a net loss in abundance of the other two (non-target) species (Figure 4). The smallest

694 losses take place under the curlew metric, with NNL of curlew, a net gain of 21 oystercatchers and a 695 net loss of 14 lapwings. The greatest loss in abundance is under the oystercatcher metric, with a net- 696 loss of 3800 lapwings and curlews combined. This result highlights the differences in habitat 697 preferences of the three species and showcases the complexity and difficulty in defining an offset credit 698 to maintain no net loss of multiple species.

4500

4000

3500

3000

2500

2000 offset offset trading 1500

1000

500 Net Net losses in species abundance as result of

0 Curlew as the NNL policy target Lapwing as the NNL policy Oystercatcher as the NNL policy target target 699

700 Figure 4: A comparison of the net losses of the non-target species across the three ecological metrics

701

702 5.2 Market Size Effects

703 The second design parameter we are interested in is how changes to the size of the market, in terms of 704 the number of landowners within a region, affects the market equilibrium and subsequent economic 705 surplus. We test this by dividing the original case study area into the three sub-markets, known as 706 Planning Areas (PA) 1, 2 and 3, which broadly align to the Local Authority planning areas found within 707 the case study region. For this analysis, we focus solely on NNL of oystercatcher as the policy target. 708 The model using the oystercatcher policy target was thus re-run for each PAs individually as 3 distinct 709 and independent sub-markets.

710 The first result of the sub-market models showed that market equilibrium was reached in each market 711 and there was no net loss of the oystercatcher policy target in all the sub-markets. Under the full market 712 region, an oystercatcher offset cost £21,978 per bird at market equilibrium. In contrast, under the sub- 713 market analysis, PA 1 had the lowest cost offset at £771 per oystercatcher, distinctly cheaper than the 714 full market-clearing price. Within PA 1 there were two land parcels that provide a high number of

715 “cheap” oystercatcher offset credits (35 in total) which are sold to 74 developers. In addition, a high 716 proportion of development takes place on land parcels which do not require offset credits (33 parcels). 717 The economic surplus for the offset suppliers was approximately 10% of the economic surplus of the 718 developers (£300,000). Comparing development in the full market with the development in PA 1, we 719 can see more development takes place in PA 1, compared with the same area in the full market (Figure 720 5).

721

722 Figure 5: A comparison of offset supply sites for oystercatchers and development impacts on oystercatcher 723 abundance under the full market scenario (left panel) and sub-market scenario (right panel). Planning 724 boundaries are shown on both maps for ease of comparison.

725 PA 2 had the highest market-clearing price for the oystercatcher offsets at £39,632 per bird. 32 land 726 parcels changed land management practice to become oystercatcher supply sites, which in total supplied 727 41 oystercatcher credits (on average 1.3 oystercatchers supplied per parcel). This is significantly smaller 728 than the number of offsets supplied per land parcel compared to Planning Areas 1 and 3 (on average 729 17.5 per parcel and 8.1 per parcel respectively). Compared to the full market model, there is 730 significantly less development taking place in PA 2, as a result of the increase in the price of an 731 oystercatcher credit.

732 PA 3 had the greatest number of land parcels requiring an offset (302), with a market-clearing price of 733 £26,699 per oystercatcher credit. The distribution of housing developments in PA3 is broadly similar 734 to the full market. This similarity is a result of PA 3 containing the majority of the land parcels which 735 supply offsets in the full market.

736 Comparing economic surplus for the developers and offset suppliers, the sub-market scenario offers a 737 small increase in the total surplus for offset suppliers of approximately £8000. This is due to the increase 738 in the additional land parcels supplying offsets in PA 3 and the increase in market clearing price for Pas 739 2 and 3 compared to the full market scenario. Surplus for housing developers reduces by £5,000 (Error! R 740 eference source not found.).

260 Full Market: Market E = £21, 979 28 Economic surplus for developers = £12,400,000 661 Economic surplus for farmers = £480,240

35 Plan area 1: Market E = £771 2 Economic surplus for developers = £364,344 74 Economic surplus for farmers = £380

41 Plan area 2: Market E = £39,632 32 117 Economic surplus for developers = £1,08,710 Economic surplus for farmers = £222,519 114 Plan area 3: Market E = £26,699 14 Economic surplus for developers = £6,150,068 302 Economic surplus for farmers = £266,077 0 200 400 600 800 Oystercatcher offsets supplied No. of suppliers No. of developers requiring offsets 741

742 Figure 6: A comparison of the number of oystercatcher credits supplied, the number of land parcels 743 becoming offset suppliers and the number of developers requiring offsets, under the full market and three 744 sub-markets models. The market equilibrium credit price (Market E) is provided for the full market and 745 three sub-markets.

746 From an ecological viewpoint, the sub-market scenario is marginally more beneficial to the non-target 747 species (curlews and lapwings) compared to the full market scenario (Figure 7). Net losses of both these 748 species are lower under the sub-market scenario, mainly as a result of the geographical re-distribution 749 of offset trades.

750

751

752

753 Figure 7: A comparison of the losses (top chart) and gains (bottom chart) in the abundance of the NNL 754 target species (oystercatcher), as well as curlews and lapwings (the unintended consequences of trading) 755 across the full market and regional sub-market scenario.

756

757

758 6 Discussion

759 Tradeable offset credits offer a promising approach to tackling the conflicts between land development 760 and biodiversity conservation. Whilst there has been a growing policy interest in the concept, and a 761 large ecological literature studying the problem, there have been relatively few contributions from 762 economists to thinking about how best to design such offset markets. Our paper is designed to make a 763 contribution to the gap in the literature. One key contribution of our paper is the development of 764 spatially explicit supply and demand curves for biodiversity offsets, which allows us to test how two 765 changes to two fundamental design parameters – the ecological target and the geographic size of the 766 market – affect the functioning of a biodiversity offset market. To do this, we develop and integrate 767 ecological and economic models which encapsulate the spatial heterogeneity in the biodiversity 768 potential of individual sites as well as spatial heterogeneity in supply and demand prices.

769 Results show that the choice of the ecological metric has a significant effect on the number of trades 770 taking place, the amount of allowable development, unintended ecological consequences, and economic 771 welfare. Each of the three species considered as NNL targets favour different land parcels, as predicted 772 by the ecological model. From a supply perspective this means that each ecological target has a varying 773 opportunity cost in terms of agricultural land foregone. From a demand perspective, land parcels with 774 development potential have varying permit requirements depending on which species is chosen as the 775 NNL target. Using the spatially derived supply and demand curves reveals that the market does not 776 always react according to theory. Here our NNL curlew target with the lowest equilibrium offset price 777 at £9,972 per curlew resulted in the fewest number of trades taking place (10 trades and 108 land parcels 778 developed). In contrast, the NNL target which resulted in the costliest equilibrium offset price (£21,979 779 per oystercatcher) had the highest number of trades (260) with 694 land parcels developed. Based on 780 our comparative statics’ analysis, we might have presumed that the species which offered the cheapest 781 offsets would have led to the most development taking place as more developers would have been 782 willing to purchase a cheaper offset. This follows from an assumption that cheaper credit prices would 783 reduce opportunity costs for developers. However, based on our integrated modelling this result is more 784 nuanced: the greatest abundance of curlews and lapwings are found in the areas with the highest 785 development values, and subsequently a greater number of credits is required by the developer per land

786 parcel developed. As such, developers are willing to pay less per offset credit. Moreover, the most 787 favourable areas for lapwing and curlew supply are in areas where the value of agriculture is higher 788 (i.e., the suppliers' opportunity cost is higher), increasing the individual farmers' unit cost of offset 789 provision. There are very few farmers who offer offsets to the market below the maximum willingness 790 to pay of the housing developer. The market is therefore “constrained,” imposing a downward pressure 791 on the offset price to a level below what many suppliers would be willing to accept.

792 In contrast, oystercatchers are recorded at lower abundances in the areas of high housing value, meaning 793 developers are required to spend less on offsetting per parcel, increasing their maximum willingness to 794 pay for a single offset and thus encouraging more farmers to supply oystercatcher offsets. In addition, 795 under the full market size scenario, all oystercatcher offsets are created at the coastal margin where the 796 value of agricultural land is low but the potential for the number of new offsets to be generated is high. 797 This influences the farmers’ willingness to accept switching from agriculture to offset generation.

798 Related to this first result is the impact of the choice of offset metric on the other two bird (non-target) 799 species considered. Whilst for each model we focus on no net loss of the target species and this condition 800 is always met, we can analyse the impact on the other species because of spatial changes in land use. 801 Across all three metrics we see a net loss in the other two bird species, with the highest losses for 802 lapwings and curlews under the oystercatcher metric. This highlights that oystercatchers have very 803 different habitat requirements compared to curlews and lapwings. Developing a market that would 804 ensure no net loss of these three species simultaneously would, therefore, be more complex and likely 805 depress the level of new housing development, although this would be ecologically more beneficial. It 806 adds further evidence to the idea that designing credit currencies is inherently complex. Simple single 807 species metrics lend themselves to simple trading rules but can result in negative impacts on other 808 measures of biodiversity.

809 In terms of economic welfare, there is a systematically large difference between consumer and producer 810 surplus changes identified in the model outcomes. Consumer surplus is consistently greater across the 811 three ecological metrics, indicating that the market benefits housing developers more than farmers. This 812 has an interesting relevance given the current state of biodiversity offset discussions within the UK 813 policy context. The current plans are backed heavily by housing developers, with many even keen to 814 adopt a “net gain agenda”. However, there is a significant lack of farmers/landowners willing to take 815 part in the planning process or engage in the biodiversity offset pilot schemes (Sullivan and Hannis 816 2015). At present, the current incentives offered by biodiversity offsetting are not strong enough for 817 these entities to begin to engage with the policy process.

818 A second design issue relates to the size of the market. In general, more efficiently functioning markets 819 are those featuring a high number of participants that stimulate trading and enhance market liquidity. 820 Such markets would likely feature less volatile prices than thin markets, reducing price uncertainty in a 821 manner conducive to more investment in conservation (Wissel and Wätzold, 2010). Our theoretical 822 model predicts that an increase in the number of sellers would increase the amount of hectares that 823 generate offsets, resulting in a downward pressure on the equilibrium credit price. In contrast, the 824 equilibrium price would increase if demand would be enhanced through an increase in the number of 825 buyers. In our empirical model we can only adjust the number of potential sellers by changing the

826 market boundaries. However, by doing this we change the spatial heterogeneity embedded in both the 827 supply and demand curves. This leads to a counter-intuitive empirical result that, when we run our 828 analysis over smaller sub-markets. These sub-markets are based broadly on Local Authority planning 829 regions within the case study. The smallest sub-market (Planning area 1) had the lowest equilibrium 830 price per unit of one offset credit compared to when trading is allowed across the full region. This 831 specific sub-market only contains two offset supply sites, which, however, offer a high number of 832 credits due to the high ecological productivity of the two land parcels (that is, changing land 833 management to grassland produces a large predicted rise in target species abundance). Since credits are 834 supplied relatively cheaply due to the low opportunity costs of the land, the total amount of development 835 increases within this area (for new development not requiring offsets) due to spatial heterogeneity in 836 land value for development. That is, land value in the sub-region is low, thus depressing the developers’ 837 maximum willingness to pay for an offset. As such, in the full market scenario where offsets are more 838 costly, development would not take place in the same locations, since developers would instead choose 839 more profitable land parcels. In contrast, for Planning areas 2 and 3, the credit price increases compared 840 to the single catchment-wide market equilibrium price of £21,000 (£26,000 and £29,000 per offset 841 respectively. In these two sub-markets where the credit price increases, demand for new development 842 falls, consistent with the theoretical prediction. Spatially this is due to these areas no longer having 843 access to the cheapest offset supply sites. This is particularly evident in Planning area 3 where new 844 offset suppliers enter the market as a result of an increase in the offset supply price.

845 Ecologically it appears that configuring smaller sub-markets is beneficial as the losses in numbers of 846 the non-target wading birds are smaller than the full market scenario. This is a result of the decline in 847 new development due to the increase in credit price in two of the sub-markets. A concern ecologically 848 under the full market scenario when oystercatchers are the NNL target species is the difference in the 849 location of habitat being lost due to development compared to habitat being created for offsets. 850 Development takes place in upland regions where oystercatchers breed in the spring and summer 851 months, but offset sites are being created along the coastal margin, a favourable overwintering habitat. 852 A benefit of the sub-market scenario then is that it forces the creation of more expensive offset sites 853 inland. If a full market scenario was preferred, a trading rule where only spring /summer habitat could 854 be traded for newly created spring /summer habitat could be specified. This likely reduces the total 855 amount of new development due to higher credit prices.

856 We would also like to note some limitations in the model and its application. Firstly, our model does 857 not consider intertemporal dynamics regarding the ecological and economic aspects of the market. In 858 our modelling approach we chose to model two states: current land use where no offsetting took place 859 and a second state where all offsetting takes place up to where the market reaches an equilibrium. From 860 an ecological viewpoint this does not consider dynamics and time lags in the generation of the offset

861 credits related to the target species, which is a key concern of ecologists regarding offsetting (Gibbons 862 and Lindenmayer 2007; Maron et al. 2012). As recommended in Needham et al (2019), if offset markets 863 are to be developed then we would recommend the use of “banked credits” where trading can only take 864 place once the credits have been certified as providing an increase in the target species, which seeks to 865 overcome some of the problems associated with uncertainty in restoration. From an economics 866 perspective, we do not consider dynamics associated with land value changes associated with new 867 housing development. The rental value of land may decrease once a threshold of development on 868 neighbouring land parcels has been reached, thus reducing potential profits for developers and reducing 869 the requirements for credits. We also assume that once land has been secured for through an offset it 870 cannot be changed into a different land use, i.e., it is protected in perpetuity. Evidence from the UK 871 biodiversity offset pilots show that landowners are strongly opposed to this policy, preferring a 10 year 872 time horizon (Hannis and Sullivan 2015). This would likely lead to NNL of the ecological target over 873 time and undermines the potential of offset markets to deliver biodiversity protection. Stronger 874 incentives would need to be offered to landowners to encourage them to set aside their land for offsetting 875 permanently.

876 The supply prices of offsets were determined solely by agricultural gross margin data. We acknowledge 877 that this approach does not consider the social implications of offset developments, for example, loss 878 of locally important offset areas and access to green spaces. This is an aspect of offsetting that is being 879 discussed in greater detail in the literature, particularly with respect to offsetting in developing countries 880 (Griffiths et al. 2019). We also assumed that trading was mandatory across the whole case study region, 881 with any development impact on the ecological metric requiring an offset. This is in contrast to current 882 offset schemes globally where the purchasing of offsets is often undertaken voluntarily by developers, 883 who can also choose to undertake offsetting in the form of compensatory mitigation directly rather than 884 purchasing offsets through a market. This has been shown to reduce the demand for third-party 885 generated offsets. This not only suppresses the development of an offset trading scheme, but in many 886 cases also leads to lower quality restoration actions than those which would have been undertaken 887 through a regulated offset supplier such as an offset bank (Bonnie 1999; Bekessy et al. 2010).

888 The analytical framework here could be used to further study various design parameters within 889 biodiversity offset markets. For example, different ecological metrics could be used, such as no net loss 890 of habitats or an overall measure of species richness; temporal relationships could be explored through 891 the use of a dynamic model that allows for feedbacks in the system to account for impacts on 892 surrounding land parcels when adjacent parcels are either developed or certain land management 893 practices are changed. Finally, rather than considering land market values alone, other associated land 894 values could be considered too, such as accounting for carbon storage potential in upland habitats or 895 recreational values associated with green spaces near developed areas.

896 7. Conclusions

897 This study provides a novel analysis of biodiversity offset markets, integrating ecological species 898 abundance modelling with economic modelling of market outcomes. Our paper has provided evidence 899 on the importance of choice of the ecological metric for trading, as determined by the policy target, and 900 of the size of the market for trading, in terms of the development of a biodiversity offset market. 901 Empirical modelling of such offset markets has shown that altering these two design parameters does 902 not necessarily affect the market in expected ways. The choice of metric affects both the scale and 903 distribution of new development, as well as the ecological gains and losses, which differed significantly 904 between our three NNL target species (we focus on wading birds). This adds further evidence to the 905 argument that defining a credit for biodiversity offset trading is inherently complex. In addition, the 906 cheapest offset credit lead to the lowest amount of new development, and conversely the most expensive 907 credits were associated with the greatest amount of new development.

908 Regarding market size, we show that a larger geographic region for trading benefits the developers. 909 This scale allows each developer to access the cheapest offset supply sites. Dividing the region into 910 three smaller sub-markets constrains developers to access to the cheapest offsets within their sub-market 911 and places downward pressure on new development in two of the three sub-markets. However, dividing 912 a market into sub-markets also has ecological implications for both target and non-target species, and a 913 policy designer would need to balance these against the economic impacts of creating sub-markets in 914 deciding whether and how to segment an offset market,

915 In conclusion, our work has provided evidence to help understand how the choice of the ecological 916 target and the size of the market affect the development and functioning of a biodiversity offset market. 917 Implementing such markets could assist in reducing development/conservation conflicts. However, 918 offsetting alone is not sufficient to guarantee no declines in wider measures of biodiversity, and existing 919 national and international biodiversity protection must remain in place (such as Special Protection Areas, 920 for example). Instead, offsetting has the potential to reduce the loss of biodiversity for ecologically 921 valuable habitats that are not within the internationally protected regions, with the aim of moving 922 towards a net gain in biodiversity nation-wide.

923

924 8. References

925 Baker, William; Sheate, William; Bennett, Teresa; Payne, David; Tucker, Graham; White, Owen; 926 Forrest, Steven. 2016. “Programme Evaluation of the Biodiversity Offsetting Pilot Programme 927 Final Report.” https://doi.org/10.13140/RG.2.1.4437.4168.

928 Barker, Nicole K. S., Stuart M. Slattery, Marcel Darveau, and Steve G. Cumming. 2014. “Modeling 929 Distribution and Abundance of Multiple Species: Different Pooling Strategies Produce Similar 930 Results.” Ecosphere 5 (12): art158. https://doi.org/10.1890/ES14-00256.1.

931 Beattie, Alastair. 2019. The Farm Management Handbook 2019/20. SRUC. www.fas.scot.

932 Bekessy, Sarah A., Brendan A. Wintle, David B. Lindenmayer, Michael A. Mccarthy, Mark Colyvan, 933 Mark A. Burgman, and Hugh P. Possingham. 2010. “The Biodiversity Bank Cannot Be a Lending 934 Bank.” Conservation Letters 3 (3): 151–58. https://doi.org/10.1111/j.1755-263X.2010.00110.x.

935 Bendor, Todd. 2009. “A Dynamic Analysis of the Wetland Mitigation Process and Its Effects on No 936 Net Loss Policy.” Landscape and Urban Planning 89 (1–2): 17–27. 937 https://doi.org/10.1016/j.landurbplan.2008.09.003.

938 BenDor, Todd K., Tianshu Guo, and Andrew J. Yates. 2014. “Optimal Advanced Credit Releases in 939 Ecosystem Service Markets.” Environmental Management 53 (3): 496–509. 940 https://doi.org/10.1007/s00267-013-0219-1.

941 Bonds, Matthew H., And Jeffrey J. Pompe. 2003. “Calculating Wetland Mitigation Banking Credits: 942 Adjusting for Wetland Function and Location.” Natural Resources Journal. Regents of the 943 University of New Mexico on behalf of its School of Law. https://doi.org/10.2307/24888894.

944 Bonnie, Robert. 1999. “Endangered Species Mitigation Banking: Promoting Recovery through Habitat 945 Conservation Planning under the Endangered Species Act.” Science of the Total Environment 240 946 (1–3): 11–19. https://doi.org/10.1016/S0048-9697(99)00315-0.

947 Brotons, Lluís, Wilfried Thuiller, Miguel B Araújo, Alexandre H Hirzel Brotons, Araú Jo, M B Hirzel, 948 and / W Thuiller. 2004. “Presence-Absence versus Presence-Only Modelling Methods for 949 Predicting Bird Habitat Suitability.” ECOGRAPHY 27: 4.

950 Bull, Joseph W., K. Blake Suttle, Ascelin Gordon, Navinder J. Singh, and E. J. Milner-Gulland. 2013. 951 “Biodiversity Offsets in Theory and Practice.” ORYX. 952 https://doi.org/10.1017/S003060531200172X.

953 Burgin, Shelley. 2008. “BioBanking: An Environmental Scientist’s View of the Role of Biodiversity 954 Banking Offsets in Conservation.” Biodiversity and Conservation. 955 https://doi.org/10.1007/s10531-008-9319-2.

956 Coralie, Calvet, Ollivier Guillaume, and Napoleone Claude. 2015. “Tracking the Origins and 957 Development of Biodiversity Offsetting in Academic Research and Its Implications for 958 Conservation: A Review.” Biological Conservation. Elsevier Ltd. 959 https://doi.org/10.1016/j.biocon.2015.08.036.

960 Stockton-on-Tees Borough Council, 2016. “Economic Growth Plan.” 961 http://egenda.stockton.gov.uk/aksstockton/images/att30085.pdf.

962 Defra. 2012. “Technical Paper: The Metric for the Biodiversity Offsetting Pilot in England.” 963 https://www.gov.uk/government/publications/technical-paper-the-metric-for-the-biodiversity- 964 offsetting-pilot-in-england.

965 Doyle, Martin W., and Andrew J. Yates. 2010. “Stream Ecosystem Service Markets under No-Net-Loss 966 Regulation.” Ecological Economics 69 (4): 820–27. 967 https://doi.org/10.1016/j.ecolecon.2009.10.006.

968 Drechsler, Martin, and Florian Hartig. 2011. “Conserving Biodiversity with Tradable Permits under 969 Changing Conservation Costs and Habitat Restoration Time Lags.” Ecological Economics 70 (3): 970 533–41. https://doi.org/10.1016/j.ecolecon.2010.10.004.

971 Dupont, Valérie. 2017. “Biodiversity Offsets in NSW Australia: The Biobanking Scheme versus 972 Negotiated Offsets in Urban Areas.” Journal of Environmental Law, January, eqw031. 973 https://doi.org/10.1093/jel/eqw031.

974 Evans, D M, R Altwegg, T W J Garner, M E Gompper, I J Gordon, J A Johnson, N Pettorelli, and 975 Darren M Evans. 2014. “Biodiversity Offsetting: What Are the Challenges, Opportunities and 976 Research Priorities for Animal Conservation?” https://doi.org/10.1111/acv.12173.

977 Fernandez, Linda, and Larry Karp. 1998. “Restoring Wetlands through Wetlands Mitigation Banks.” 978 Environmental and Resource Economics 12 (3): 323–44. 979 https://doi.org/10.1023/A:1008225021746.

980 Ferrari, Edward, and Karl Dalgleish. 2019. “Tees Valley Local Housing Markets.” 981 https://www4.shu.ac.uk/research/cresr/sites/shu.ac.uk/files/tees-valley-local-housing- 982 markets.pdf

983 Franks, Samantha E., David J.T. Douglas, Simon Gillings, and James W. Pearce-Higgins. 2017. 984 “Environmental Correlates of Breeding Abundance and Population Change of Eurasian Curlew 985 Numenius Arquata in Britain.” Bird Study 64 (3): 393–409. 986 https://doi.org/10.1080/00063657.2017.1359233.

987 Gibbons, Philip, and David B. Lindenmayer. 2007. “Offsets for Land Clearing: No Net Loss or the Tail 988 Wagging the Dog?: Comment.” Ecological Management and Restoration. 989 https://doi.org/10.1111/j.1442-8903.2007.00328.x.

990 HM Government 2018. “Land Value Estimates.” 2018. 991 https://www.gov.uk/government/collections/land-value-estimates.

992 Griffiths, Victoria F., Joseph W. Bull, Julia Baker, and E. J. Milner-Gulland. 2019. “No Net Loss for 993 People and Biodiversity.” Conservation Biology 33 (1): 76–87. 994 https://doi.org/10.1111/cobi.13184.

995 Hannis, Mike, and Sian Sullivan. n.d. Offsetting Nature? Habitat Banking and Biodiversity Offsets in 996 the English Land Use Planning System. Accessed February 3, 2020. 997 http://www.greenhousethinktank.org.

998 Hayhow, DB, Eaton, MA, Stanbury, AJ, Burns, F, Kirby, WB, Bailey, N, Beckmann, B, Bedford, J, 999 Boersch-Supan, PH, Coomber, F, Dennis, EB, Dolman, SJ, Dunn, E, Hall, J, Harrower, C, Hatfield, 1000 JH, Hawley, J, Haysom, K, Hughes, J, Johns, DG, Mathews, F, McQua, N. 2019. “State of Nature 1001 2019.” https://nbn.org.uk/wp-content/uploads/2019/09/State-of-Nature-2019-UK-full-report.pdf.

1002 Heal, Geoffrey M. 2005. “Arbitrage, Options and Endangered Species.” SSRN Electronic Journal, 1003 November. https://doi.org/10.2139/ssrn.457330.

1004 Hough, Palmer, and Morgan Robertson. 2009. “Mitigation under Section 404 of the Clean Water Act: 1005 Where It Comes from, What It Means.” Wetlands Ecology and Management 17 (1): 15–33. 1006 https://doi.org/10.1007/s11273-008-9093-7.

1007 JNCC, Joint nature Conservation Committee). 2020. “Special Protection Areas – Overview | JNCC - 1008 Adviser to Government on Nature Conservation.” Special Protection Areas – Overview. 2020. 1009 https://jncc.gov.uk/our-work/special-protection-areas-overview/.

1010 Kangas, Johanna, and Markku Ollikainen. 2019. “Economic Insights in Ecological Compensations: 1011 Market Analysis With an Empirical Application to the Finnish Economy.” Ecological Economics 1012 159 (May): 54–67. https://doi.org/10.1016/j.ecolecon.2019.01.003.

1013 Laitila, Jussi, Atte Moilanen, and Federico M. Pouzols. 2014. “A Method for Calculating Minimum 1014 Biodiversity Offset Multipliers Accounting for Time Discounting, Additionality and Permanence.” 1015 Methods in Ecology and Evolution 5 (11): 1247–54. https://doi.org/10.1111/2041-210X.12287.

1016 Lapeyre, Renaud, Géraldine Froger, and Marie Hrabanski. 2015. “Biodiversity Offsets as Market-Based 1017 Instruments for Ecosystem Services? From Discourses to Practices.” 1018 https://doi.org/10.1016/j.ecoser.2014.10.010.

1019 Lockhart, Andy. 2014. “Developing an Offsetting Programme: Tensions, Dilemmas and Difficulties in 1020 Biodiversity Market-Making in England.” Environmental Conservation 42 (4): 335–44. 1021 https://doi.org/10.1017/S0376892915000193.

1022 Maron, Martine, Richard J. Hobbs, Atte Moilanen, Jeffrey W. Matthews, Kimberly Christie, Toby A. 1023 Gardner, David A. Keith, David B. Lindenmayer, and Clive A. McAlpine. 2012. “Faustian 1024 Bargains? Restoration Realities in the Context of Biodiversity Offset Policies.” Biological 1025 Conservation 155 (October): 141–48. https://doi.org/10.1016/j.biocon.2012.06.003.

1026 McKenney, Bruce A., and Joseph M. Kiesecker. 2010. “Policy Development for Biodiversity Offsets: 1027 A Review of Offset Frameworks.” Environmental Management. https://doi.org/10.1007/s00267- 1028 009-9396-3.

1029 Needham, Katherine, Frans P de Vries, Paul R Armsworth, and Nick Hanley. 2019. “Designing Markets 1030 for Biodiversity Offsets: Lessons from Tradable Pollution Permits.” Journal of Applied Ecology 1031 56 (6): 1429–35. https://doi.org/10.1111/1365-2664.13372.

1032 Office of Heritage, NSW. 2014. “Biodiversity Legislation Review OEH Paper 4: Conservation in Land- 1033 Use Planning.” In . www.environment.nsw.gov.au.

1034 Potts, Joanne M., and Jane Elith. 2006. “Comparing Species Abundance Models.” Ecological Modelling 1035 199 (2): 153–63. https://doi.org/10.1016/j.ecolmodel.2006.05.025.

1036 Quétier, Fabien, Baptiste Regnery, and Harold Levrel. 2014. “No Net Loss of Biodiversity or Paper 1037 Offsets? A Critical Review of the French No Net Loss Policy.” Environmental Science and Policy 1038 38 (April): 120–31. https://doi.org/10.1016/j.envsci.2013.11.009.

1039 Robertson, Morgan M. 2004. “The Neoliberalization of Ecosystem Services: Wetland Mitigation 1040 Banking and Problems in Environmental Governance.” Geoforum 35 (3): 361–73. 1041 https://doi.org/10.1016/j.geoforum.2003.06.002.

1042 Scanlon, Jane. 2007. “An Appraisal of the NSW Biobanking Scheme to Promote the Goal of Sustainable

1043 Development in NSW.” Journal of International and Comparative Environmental Law. 2007. 1044 http://classic.austlii.edu.au/au/journals/MqJlICEnvLaw/2007/4.html.

1045 Simpson, Katherine. 2011. “Implementing an Ecosystem Approach:The Case of the Teesmouth and 1046 Cleveland Coast European Marine Site.”

1047 Smith, Ken. 2011. “The State Of The Natural Environment Of The Tees Estuary A Review Of The Bird 1048 Chapter.” http://www.inca.uk.com/wp-content/uploads/2012/01/BIRDS-SONET-UPDATE- 1049 2011.pdf.

1050 Sullivan, S., and M. Hannis. 2015. “Nets and Frames, Losses and Gains: Value Struggles in 1051 Engagements with Biodiversity Offsetting Policy in England.” Ecosystem Services 15 (October): 1052 162–73. https://doi.org/10.1016/j.ecoser.2015.01.009.

1053 Tattoni, Clara, Franco Rizzolli, and Paolo Pedrini. 2012. “Can LiDAR Data Improve Bird Habitat 1054 Suitability Models?” Ecological Modelling 245 (October): 103–10. 1055 https://doi.org/10.1016/j.ecolmodel.2012.03.020.

1056 Ten, Kerry, Kate Jo Treweek, and Jon Ekstrom. 2010. “The Use Of Market-Based Instruments For 1057 Biodiversity Protection-The Case Of Habitat Banking European Commission DG Environment.” 1058 www.eftec.co.uk.

1059 Tucker, Graham, Allen, Ben, Conway, Mavourneen, Dickie, Ian, Hart, Kaley, Rayment, Matt, Schulp, 1060 Catharina, van Teeffelen, Astrid. 2013. “Policy Options for an EU No Net Loss Initiative.” 1061 www.ieep.eu.

1062 US Army Corps of, Engineers. 2020. “Regulatory In-Lieu Fee and Bank Information Tracking System.” 1063 Online. 2020. https://ribits.usace.army.mil/ribits_apex/f?p=107:2.

1064 Venables, W N, and B D Ripley Springer. n.d. “Modern Applied Statistics with S Fourth Edition.” 1065 Accessed February 3, 2020. http://www.insightful.com.

1066 Vries, Frans P. de, and Nick Hanley. 2016. “Incentive-Based Policy Design for Pollution Control and 1067 Biodiversity Conservation: A Review.” Environmental and Resource Economics. Springer 1068 Netherlands. https://doi.org/10.1007/s10640-015-9996-8.

1069 William Bull, Joseph, and Niels Strange. n.d. “The Global Extent of Biodiversity Offset Implementation 1070 under No Net Loss Policies.” Nature Sustainability. Accessed December 16, 2019. 1071 https://doi.org/10.1038/s41893-018-0176-z.

1072 Wissel, Silvia, and Frank Wätzold. 2010. “A Conceptual Analysis of the Application of Tradable 1073 Permits to Biodiversity Conservation” Conservation Biology 24 (2): 404–11. 1074 https://doi.org/10.1111/j.1523-1739.2009.01444.x.

1075

1076

1077

1078

1079 7 Supplementary material

1080

Figure A: Flow chart of the integrated ecological – economic model to determine the market-clearing price for biodiversity offsets under three different policy targets 1081

1082

1083 7.1 Data sources

1084 Land classification

1085 For each 1 by 1km land parcel, ESRI ArcGIS v10.4 (ESRI, Redlands, CA) was used to derive 1086 information on the UK Biodiversity Action Plan Broad Habitats classes and crop cover. Land is 1087 classified using the Centre of Ecology and Hydrology (CEH) Land Cover Map 2015 (LCM2015) and 1088 CEH Land Cover plus Crops 2017 (Rowland et al 2015). LCM2015 is created by classifying two-date 1089 composite images and is based mainly on data from Landsat-8 (30m resolution) supplemented with 1090 AWIFS data (60 m resolution) as required (NERC CEH). This produces 21 land classifications 1091 following the UK Biodiversity Action Plan Broad Habitats classes including urban, improved grassland, 1092 arable and horticulture. A limitation for our work is that the data is periodically updated (the previous 1093 iteration being LCM2007) and as such there may be differences between the actual habitats identified 1094 in 2015 and our model year (2017). To derive crop coverage, we used the 2017 CEH Land Cover plus 1095 Crops map. This is the first high-resolution map of UK crops and classifies two million land parcels 1096 based on the LCM2015. Crop data is derived from a combination of Copernicus Sentinel-1 C-band SAR 1097 (Synthetic Aperture Radar) and Sentinel-2 optical data. Accuracy of the data is assessed using crop type 1098 data collated by the Rural Payments Agency, Rural Payments Wales and Scottish Government Rural 1099 Payments and Services which detail all land registered for agricultural use and are submitted annually 1100 by farmers. There are 11 crop types (beet, field beans, grassland, maize, oilseed rape, potatoes, spring 1101 barley, winter barley, spring wheat, winter wheat (including oats) and other. The LCM2015 and Land 1102 Cover plus Crops 2017 datasets were combined using the Tabulate Intersection spatial statistics tool in 1103 ArcGIS. This derived the total area for each habitat and crop type within the 1 by 1 km land parcel. In 1104 over 95% of land parcels the Land Cover plus Crops data aligned with the LCM2015 arable and 1105 horticulture classification. Where differences were identified, for example, winter wheat in the Land 1106 Cover plus Crops dataset overlaying a layer containing grassland from the LCM 2015 data, we assumed 1107 land use from the Land Cover plus Crops dataset. There may be disagreement between the two datasets 1108 based on the changes to field boundaries between the collection years (limited impact) but is more likely 1109 due to changes in validation methods for the data between the years 2015, 2016 and 2017 as 1110 acknowledged by CEH.

1111

1112

1113

1114

1115 Agricultural profits and land use data

1116 The value of agricultural land was calculated using gross margin data for crops and livestock from the 1117 SRUC Farm Management Handbook. Gross margin is a widely-used measure of net benefit from crop 1118 or livestock choice in farm accounting, being constructed from total revenues from sale of 1119 crop/livestock product minus the variable costs of production. For crop data, the crop coverage derived 1120 from the Land Cover plus Crops dataset was combined with the crop gross margin data from the SRUC 1121 Farm Management Handbook. Gross margin values are given for each crop type per ha on a productivity 1122 range from low, medium to high yield. To account for differences in yield, data on soil quality at the 1123 1km2 resolution was derived and linked to the crop distribution.

1124 The calculation of livestock gross margins were more complex. Data on livestock is sourced from the 1125 England Agricultural Census conducted in June each year. This surveys farmers via a postal 1126 questionnaire with the farmer declaring agricultural activity on their land. Data for non-surveyed farms 1127 is extrapolated from previous years and trends on comparable farms. This data is made available at the 1128 5 by 5 km grid square resolution and lists the number of livestock by type within the 5 km grid square. 1129 The most recent release is the 2010 data. Using the livestock gross margin data from the SRUC Farm 1130 Management Handbook, the total gross margin for all livestock within the 5 km2 grid square was 1131 calculated. Using the Tabulate Intersection Tool within ArcGIS, livestock gross margin data on the 5 1132 km resolution was then disaggregated to the 1 by 1 km land parcel resolution by attributing these values 1133 to only those habitats which would likely contain livestock: improved grassland, semi natural grasslands 1134 and heather grassland. We recognise that there are some limitations with this disaggregation approach, 1135 for example, it does not fully take into account differences in productivity between the various grassland 1136 types, however, given the scale of the case study region we believe this still creates a reliable indication 1137 for livestock gross margins.

1138 Land values for housing development

1139 To establish demand for offset credits data is needed on the potential value of land for development. 1140 Ideally, we would secure value for residential land within each parcel. However, due to data availability 1141 restrictions, we used a proxy measurement for the potential value of land for residential housing using 1142 the UK Government’s price paid dataset. HM Land Registry publishes transaction data for all residential 1143 properties sold from 1995 onwards and this is made available at the most detailed postcode level for 1144 England. To derive a land value proxy, property prices were first converted in 2017 prices using the 1145 Bank of England’s RPI figures (this is the preferred inflation measure for house prices within the UK

1146 as used in the House Price Index6). For price paid data of below £30,000 and above £400,000 additional 1147 screening was undertaken to ensure the values were representative of the postcode region. Duplicate 1148 sales of the same property were identified and only the most recent transaction used in calculations. 1149 The average house sold price for each postcode sector was then calculated. This data was overlaid with 1150 the 1 x 1 km land parcels in ArcGIS. This allowed the average sold price for each land parcel to be 1151 derived based on how many postcode sectors were present within a 1km2, the area covered by these 1152 postcodes within the grid and the price paid value associated with each postcode. This calculated a 1153 weighted average value for house prices sold at the 1km2 resolution. Where there was limited 1154 transaction data for a postcode sector, a smoothing function was applied in ArcGIS to calculate a 1155 predicted value for a 1km2 parcel based on the values of the surrounding land parcels. From government 1156 and industry reports, it has been estimated that land value represents between 30% and 50% of the price 1157 paid for a residential property.

1158 Bird Data

1159 Species data is obtained from the British Trust of Ornithology (BTO) Breeding Bird Survey (BBS). 1160 BBS survey squares are randomly selected from all 1 by 1km grid squares in the UK National Grid and 1161 the aim is that the same surveys are surveyed each year. Data was received for the Eastern region of the 1162 UK for the sampling years 2016 and 2017. BBS data is collected by volunteers who conduct three field 1163 visits to each survey square: a reconnaissance visit and two bird recording visits. Surveys take place 1164 during the early breeding season (April – mid-May) and a second visit at least for weeks later (mid- 1165 May to the end of June). Counts begin early in the early morning to maximise bird activity and last 1166 approximately 90 minutes. Volunteers record all the birds they see or hear as they walk methodically 1167 along their transect routes (BTO, 2019). To analyse species abundance the maximum count of the two 1168 visits was taken in line with BBS methodology. Species that were not recorded are assigned a count 1169 value of zero. Between 18% and 23% of the grids sampled contained at least one bird.

1170 Explanatory variables were derived using a series of spatially referenced data products: habitats were 1171 described using the land classification data derived from the LCM2015 and the Land Cover plus Crops 1172 map (21 classifications). A further set of environmental variables were derived using OS Open Map 1173 products. The average altitude within the grid square was derived using OS Terrain 5 digital terrain 1174 model (DTM). OS Open Map local (vector) was used to derive the distance from the nearest railway

6 The index applies a statistical method, called a hedonic regression model, to the various sources of data on property price and attributes to produce estimates of the change in house prices each period. The index is published monthly. More detail available from: https://landregistry.data.gov.uk/app/ukhpi

1175 line, overhead pylon, road (divided into Motorway, A road and B road) watercourse and tidal boundary. 1176 Distance from the centroid of each grid square to the polygon/polyline feature was calculated using the 1177 Arc GIS analysis tool ‘Near’ within the ‘Proximity’ tool box. In addition, road density, pylon density 1178 and railway density within the grid square were derived using the ‘Line Density Tool’ within the Spatial 1179 Analyst toolbox. For road density, two measures were calculated: unweighted where all three road types 1180 were given a value of 1; and a second weighted density with Motorways weighted highest (5), A roads 1181 (3) and all other roads (1). These variables were chosen based on the existing literature for predicting 1182 bird habitat suitability.

1183 7.2 Ecological Modelling Results

1184 Table A: Predictive ecological model results for curlew, lapwing and oystercatcher species using a pooled 1185 dataset for the years 2016 and 2017.

Curlew Pooled Sample: Lapwing Pooled Oystercatcher Pooled VARIABLES GLM with Binomial Log Sample: GLM with Sample: GLM with Link Binomial Log Link Binomial Log Link

Standard Standard Standard Habitat: Coefficient Coefficient Coefficient Error Error Error Acid Grassland (area ha) 0.003 -0.011 -0.036*** -0.009 -0.023*** -0.009 Bog (area ha) 0.001 -0.011 -0.043*** -0.01 -0.024*** -0.009 Broadleaf woodland (area ha) -0.053*** -0.012 -0.081*** -0.011 -0.041*** -0.01 Calcareous grassland (area ha) 0.023 -0.017 -0.036** -0.016 0.002 -0.017 Coniferous woodland (area ha) -0.021** -0.011 -0.074*** -0.01 -0.033*** -0.009 Fen, marsh and swamp (area ha) -0.037 -0.034 -0.005 -0.013 -0.022 -0.014 Freshwater (area ha) -0.025* -0.014 -0.048*** -0.013 0.011 -0.011 Heather (area ha) 0.004 -0.011 -0.038*** -0.01 -0.023** -0.009 Heather grassland (area ha) 0.008 -0.011 -0.035*** -0.01 -0.011 -0.009 Inland rock (area ha) -0.097*** -0.027 -0.130*** -0.038 -0.053*** -0.016 Littoral rock (area ha) 0.088 -0.149 -0.646*** -0.177 0.283*** -0.093 Littoral sediment (area ha) 0.023 -0.024 -0.168*** -0.05 0.076*** -0.022 Neutral grassland (area ha) 0.048 -0.043 0.036 -0.035 -0.037 -0.048 Saltmarsh (area ha) 0.063*** -0.023 0.029* -0.017 0.047*** -0.016 Saltwater (area ha) 0.013 -0.015 -0.086*** -0.019 0.004 -0.013 Suburban (area ha) -0.079*** -0.014 -0.100*** -0.012 -0.031*** -0.009 Supralittoral rock (area ha) -0.543* -0.321 0.082 -0.159 -0.118 -0.152 Supralittoral sediment (area ha) -0.044** -0.022 -0.048*** -0.015 0.003 -0.011 Urban (area ha) -0.068*** -0.019 -0.078*** -0.015 -0.064*** -0.012 Crop type: Beets -0.072 -0.046 -0.019 -0.013 0.002 -0.014 Field Beans -0.040** -0.019 -0.039*** -0.012 -0.061*** -0.017 Improved Grassland 0.002 -0.011 0.028 -0.009 -0.01 -0.009

Maize -0.043 -0.028 -0.027* -0.015 -0.089*** -0.019 Oil Seed Rape -0.032*** -0.012 -0.048*** -0.01 -0.054*** -0.011 Potatoes 0.027 -0.026 -0.032** -0.014 0.014 -0.012 Spring Barley -0.003 -0.013 -0.049*** -0.012 0.004 -0.01 Spring Wheat -0.01 -0.013 -0.034*** -0.011 -0.006 -0.01 Winter Barley -0.024* -0.014 -0.040*** -0.012 -0.016 -0.012 Winter Wheat -0.045*** -0.013 -0.052*** -0.01 -0.042*** -0.009 other -0.033** -0.013 -0.030*** -0.01 -0.020** -0.01 Overhead Pylons Density (km2) -0.445*** -0.116 -0.336*** -0.117 -0.375** -0.155 Tidal (distance km) 0.015*** -0.003 0.013*** -0.004 -0.002 -0.003 Watercourse (distance km) -0.232** -0.091 -0.409*** -0.085 -0.401*** -0.087 Model year dummy: 0 = 2016, 1= -0.02 -0.091 0.051 -0.101 0.013 -0.103 2017 Constant 0.771 -1.053 4.350*** -0.902 1.644* -0.853 Observations 3,626 3,620 3,610 chi2 835.6 3.776 525.4 *** p<0.01, ** p<0.05, * p<0.1 1186

1187 7.3 Data Sources

1188 British Trust for Ornithology Breeding Bird Survey

1189 Many thanks to the British Trust of Ornithology for supplying 2016 and 2017 Breeding Bird Survey 1190 Data for oystercatcher, curlew and lapwing. The BTO/JNCC/RSPB Breeding Bird Survey is a 1191 partnership jointly funded by the British Trust for Ornithology (BTO), Royal Society for the Protection 1192 of Birds (RSPB) and the Joint Nature Conservation Committee (JNCC), with fieldwork conducted by 1193 volunteers.

1194 EDINA Agricultural Census

1195 England and Wales Agricultural census data supplied by EDINA at the University of Edinburgh and 1196 the Department of Environment, Food and Rural Affairs (DEFRA) for England, The Welsh Assembly 1197 Government, or The Scottish Government (formerly SEERAD).

1198 CEH Land Cover

1199 Rowland, C.S., Morton, R.D., Carrasco, L., McShane, G., O'Neil, A.W., Wood, C.M. (2017). Land 1200 Cover Map 2015 (vector, GB). NERC Environmental Information Data Centre 1201 https://doi.org/10.5285/6c6c9203-7333-4d96-88ab-78925e7a4e73

1202 CEH Land Cover® plus Crops

1205 Soil Quality - LandIS

1206 Reproduced from National Soil Resources Institute Soilscapes viewer. © Cranfield University 2019. 1207 No part of this publication may be reproduced without the express written permission of Cranfield 1208 University.

1209 HM Registry House price data

1210 Obtained from https://data.gov.uk/dataset/4c9b7641-cf73-4fd9-869a-4bfeed6d440e/hm-land-registry- 1211 price-paid-data

1212 Contains HM Land Registry data (c) Crown copyright and database right 2018.

1213 This data is licensed under the Open Government Licence v3.0.

1214 Ordnance Survey Open Data

1215 Contains OS data © Crown Copyright OS Open Roads, OS Open Rivers, OS Terrain® 50, Boundary- 1216 Line™, Code-Point® Open and 1:250 000 Scale Colour Raster, 2018