Edinburgh Explorer A new approach to assessing the risk to woodland from pest and diseases

Pests and diseases pose a growing threat to woodlands from both 11 endemic sources, and increasingly, from inter-regional transmission. Strong 12 comparative analyses of this threat are needed in order to develop preventative 13 measures. Such analyses should include estimates of the potential worst-case loss 14 from all relevant pest and disease threats to key tree species. Existing approaches 15 tend to focus on individual assessments of the risk from a single pest or disease, 16 or assessments of overall trends. Effective risk management requires more 17 comprehensive quantified assessments of the overall threat to woodland, that 18 includes comparisons of the threat to individual tree species and identification of 19 the potentially most damaging pest and diseases. Such assessments support 20 important policy and management decisions including: species selection; 21 preventative action; and the size of buffers against losses from forest carbon 22 projects. Here we present a new approach that supports a systematic, risk-based 23 assessment of the future threat to a given woodland from all known individual 24 pest and diseases, and to constituent individual tree species, based on a risk 25 management approach taken from the finance sector, but hitherto not applied in 26 an ecological context. Unknown or unidentified pests and diseases can 27 systematically be added in future as identified. We demonstrate the method 28 through a case study evaluating the threat to projects certified under the UK’s 29 Woodland Carbon Code. The approach can be adapted to any woodland resource 30 worldwide. Its novelty lies in the simplification of complex threats, from 31 numerous pests and diseases, to measures that can be used by a range of forest 32 stakeholders.


37
Pest and diseases (P&D) represent a major threat to the world's forests (Wingfield et al. 2015;38 Flower and Gonzalez-Meler 2015) yet there is a paucity of information on statistical trends on the 39 impact and rate of forest loss caused by P&D (Petrokofsky et al. 2013). Furthermore, the threat is 40 rising: increasing connectivity between markets through growth in international trade, increases the 41 volume of shipping and air transport that can act as vectors for the transmission of P&D between 42 countries, and is leading to increasing introductions of exotic P&D (Lovett et al. 2006;Wingfield et al. 43 2015). P&D can also be transmitted via infected saplings or other genetic resources (Brasier 2008;44 Garnas et al. 2012). 45 46 Ecologists, foresters, forest owners, asset managers, policymakers and investors need to assess and 47 quantify the potential risks to woodlands, including pest and disease risk, to improve operational 48 management of risk and to factor risk into business, investment and policy decisions (Guy 2006;49 Forestry Commission 2011;Defra 2014;FIM 2015). Risk management decisions are traditionally 50 based on worst-case assessments of potential losses rather than forecasts of expected losses: for 51 example, quantifying the level of loss that will not be exceeded to a 99% level of confidence, as 52 opposed to quantifying the average expected loss (Hopkin 2014). Such comparable assessments of 53 the threat to individual tree species could contribute to decisions on the risk versus return of 54 planting different species for carbon and timber purposes. Comparisons of the threat from individual 55 P&D could help prioritise research and target resources efficiently at preventative measures. These 56 assessments could also help Fforest carbon standards, such as the Verified Carbon Standard and the 57 UK's Woodland Carbon Code (WCC), must define procedures for determining how much 58 sequestered carbon should be set aside against future losses (Verified Carbon Standard 2012;59 Forestry Commission 2014). Current assessments are inadequate to provide analyses at the 60 woodland, tree species and P&D level, to support such practices as explained below. 61 62

Limitations of existing information 63 64
In terms of providing the context to this threat, significant qualitative information on P&D exists at 65 national and regional scales, but little comprehensive quantitative information is available at a global 66 scale that does not focus on individual P&D (FAO 2009;van Lierop et al. 2015). The 2010 Global 67 Forest Resources Assessment by the Food and Agricultural Organisation of the United Nations 68 estimated that in 2005, 1.6% of the world's forests were affected by insects and 0.2% were affected 69 by diseases (FAO 2010). However, in the most recent assessment (FAO 2014), countries were asked 70 to report on the most significant outbreaks, but only 75 countries out of 155 were able to report on 71 the area of forest affected by P&D or severe weather (van Lierop et al. 2015). In these countries, 72 P&D and severe weather damaged 141.6 million hectares of forest, or 5% of the total forest area 73 (van Lierop et al. 2015). Of this, 98.0 million hectares of damage was caused by P&D (van Lierop et 74 al. 2015). 75 3 76 A recent analysis found that between 1950 and 2000, living organisms accounted for 16% of the 77 total wood damaged by natural disturbances in Europe, and 8% of this was attributed to bark 78 beetles alone (Schelhaas et al. 2003). Whilst this was the first comprehensive quantitative 79 assessment of the overall historic rate of loss caused by natural disturbances (including P&D) in 80 Europe, it does not provide breakdowns of losses from individual P&D species, or the impact on 81 individual tree species. 82 83 Existing records of P&D losses in Britain are sparse. From 1987 to 2006 the Forestry Commission 84 monitored changes in forest condition through surveys, which included damage from insect and 85 fungi (Forestry Commission 1987-2006. The information provided related to the current state of the 86 crown condition from cumulative attack and did not provide information on mortality. The 2010 UK 87 submission to the FAO's Global Forest Resources Assessment (FAO, 2010) estimated impact using a 88 threshold of "cause mortality or such severe dieback that the forest ecosystem changes". Using this 89 criterion around 1,000 hectares per year was estimated to be newly affected by disturbance from 90 insects and less for other diseases, equating to significantly less than 1% of the forest area lost per 91 year. However, the Forestry Commission's most recent submission for the UK did not quantify losses 92 at all. It provided a list of recent outbreaks and insects and diseases affecting UK trees but stated 93 that 'estimates of areas affected are not directly available' (FAO 2014 individual P&D, the majority of which do not attempt to quantify the threat that they pose (Inward 116 et al.;Mitchell et al. 2014). Many quantified assessments do exist for individual P&D (Evans, Evans, 117 and Ikegami 2008;Taylor and MacLean 2008;Harwood et al. 2009;Brasier and Webber 2010;118 Chadfield and Pautasso 2012;Stadelmann et al. 2013;Straw et al. 2013;Pukkala et al. 2014;Green 119 et al. 2015) but each study uses different methodologies and timeframes and so cannot be 120 aggregated to tree species or woodland level analyses. A range of techniques can be used to assess 121 losses at the woodland resource level, but do not provide breakdowns of the risks to individual tree 122 species or identify all of the constituent individual P&D threats. These include techniques assessing 123 crown condition (Blum et al. 2015;Morin et al. 2015) linking forest scenario models to climate and 124 bark beetle outbreaks (Seidl et al. 2009); frameworks for modelling P&D impacts using tree 125 ecophysiology (Dietze and Matthes 2014); and models linking net primary production, physiology 126 and pests (Pinkard et al. 2011 representing about 0.5 million trees (Brasier and Webber 2010). Likewise, the threat from P&D for 139 which historical data has not been gathered for long enough to capture cyclical outbreaks or 140 maximum possible losses would be underestimated. 141

142
A key source of information on current threats is the network of Regional Plant Protection 143 Organisations ( To address these gaps, we present a novel methodology that derives a single measure of the threat 159 to woodland resources, and to its constituent tree species, from the consolidation of standardised 160 evaluations of individual P&Ds threats. Standardisation ensures that P&Ds are assessed over the 161 same time period and by the same approach so that assessments can be compared and aggregated. 162 Our approach was adapted from methods used in the finance sector for risk management and 163 insurance purposes. The parallel drawn here is as follows: regulators in the banking sector must 164 ensure that banks have sufficient capital set aside in reserve, to cover unexpected losses and to 165 prevent institutional failure. All bank assets -whether mortgages, loans, trading positions or other -166 are risk-weighted and aggregated to determine the amount to set aside against overall losses. (Hull 167 2012;Crouhy et al. 2014). This standardised approach provided the inspiration for us to develop our 168 P&D methodology. 169 170 Use of a standardised methodology for assessing the risk from individual assets by the finance sector 171 allows rapid assessment in the timescales necessary for policy and business decisions to be made. It 172 is separate from, but in addition to, the more detailed assessments made by bank employees when 173 setting up financial transactions with customers. Similarly, the P&D methodology described here 174 uses a standardised approach for rapid assessment of overall threats to woodland, but should be 175 used in addition to the more detailed and nuanced assessments of individual P&D risks such as Pest 176 Risk Assessments. Similarly, risk-weighted assets (the assets of a financial institution weighted by risk and used to 186 determine the amount of capital that must be set aside against potential losses) for financial risks 187 are determined by summing the risk of individual positions, each of which is calculated by 188 multiplying the probability of a risk event occurring by an estimate of its financial consequence, 189 expressed as a proportion ofin terms of a percentage loss, which would be the worst-case loss. For 190 example, credit risk (or the risk of default) for a mortgage position, would be calculated from the 191 estimated likelihood of default of the counterparty multiplied by the worst-case loss that would 192 occur if the mortgage defaulted. These risk values are then aggregated according to the level of 193 6 assessment e.g. departmental or bank level, and used to determine the amount of capital to set 194 aside against losses. (Hull 2012;Crouhy et al. 2014). 195 196 The detailed calculations for estimating probability and consequence for credit, market and 197 operational financial risks are different. Our approach to quantifying P&D risk is described here by 198 comparison to the calculation of credit risk. Risk weighted assets (RWA) for Credit risk are calculated 199 for each financial exposure of the bank (such as a mortgage or loan) and aggregated as per the 200 following Equation (1)  individual P&D for each tree species in a woodland; 208  Aggregated risk factors for each individual tree species from all P&D that threaten them; 209  An aggregated risk factor of the overall risk to a woodland resource. 210

211
The P&D risk factors identify which P&Ds pose the biggest threats from the hundreds of known P&D 212 threats. Risk factors at the tree species level provide a comparison of the relative threat to different 213 tree species, and can assist in analyses of the risk versus return of planting different species. The 214 final woodland resource risk factor provides an assessment of how much woodland is at risk. 215

216
It is important to note that risk-weighting P&D for risk management purposes is fundamentally 217 different to forecasting the expected loss from each P&D. Risk factors for individual P&D are not 218 expected to cover the potential loss from that individual P&D, in the same way that premiums for a 219 single household insurance policy will not cover the cost of rebuilding that house if it burns down. 220 However, the sum of all household premiums, are intended to cover the potential claims from all 221 insured households (Thoyts 2010). The sum of all risk factors for all P&D, taken together can help 222 determine how much might be lost at the forest scale from all P&D. Risk factors at P&D level, do 223 however provide a means of comparing the relative risk from different P&D, in the same way that 224 higher premiums relate to a higher assessment of insurance risk. 225 7 226 Risk assessment of this type uses a worst-case loss (also termed unexpected or catastrophic loss) for 227 potential loss as opposed to the most likely or expected one. For wind risk to trees, for example, this 228 would be the difference between average annual windthrow and the kind of devastation caused by a 229 severe storm. Definitions of worst-case are set according to the confidence level sought but are 230 close to 100%. For example, in finance, the market risk assessment provides a value that should not 231 be exceeded to a 99% degree of confidence. The P&D risk factor calculation is adapted from Equation (1). For credit risk, the 'probability of an 240 event' as per Hopkin's definition, is the probability of default (PD) in Equation (1). For P&D this 241 relates to the likelihood of arrival and establishment. Similarly, for credit risk, Hopkin's 242 'consequence' or impact is the loss given default (LGD) in Equation (1). For P&D this is the worst-case 243 loss that could occur within the range of the P&D. The exposure at default (EAD) in Equation (1) is 244 not required as we express the final risk factors as a proportion of tree species or woodland at risk 245 rather than a financial value. P&D risk factors are estimated for each P&D/tree species combination 246 for a given time period (t) and geographical area (g). The risk factor (RiskFac) for a pest and disease: 247 p1, affecting tree species: t1, is therefore estimated as follows: 248 249 RiskFacp1t1 = ProbArrp1 x ProbMaxp1 x ProbLossp1t1 Where ProbArrp1 = Annual probability of arrival and establishment of p1 in geographical area (g) during time period (t) expressed as a %. ProbMaxp1 = Maximum possible proportion of geographical area (g) that p1 could impact expressed as a %. ProbLossp1t1 = Worst-case loss from tree species t1, caused by p1 within time period (t) within the potential range of p1 expressed as a %.proportion.
(2) 250 251 Formatted Table   Commented [G2]: The units for this equation need to be reset here and in the rest of the paper. At present you have each term as a %, which produced a very large number with units of 'percentage cubed' that cannot be correct. I suggest you select one as a % (i.e. 89%) and set the others as proportions (i.e. 0.24 and 0.52) so when the three numbers are multiplied together you get a sensible % figure for RiskFac. Also, the equation number should be moved up. RESPONSE: changed to % for ProbArr and proportions for the rest. Plus text edited throughout. Equation number also moved up.  The calculation of ProbLoss therefore requires a different approach to estimation for each practical 281 application of this methodology, i.e. depending on the time period being assessed, age distribution 282 of the tree species, definition of loss (e.g. yield reduction or mortality), and replanting options. 283 However, the information on the fastest rate of spread of the P&D to its full range, the age of trees 284 that it affects, and the worst-case impact on yield/mortality, is the same regardless of the 285 application, and so this information need only be estimated once and can then be used in other 286 applications of the method. 287 288 An estimate of the worst-case loss from each P&D/tree species combination needs to be estimated 289 for each application. This is done by estimating the potential loss that could occur if the P&D arrived 290 in each year during the time period (t), and then expanded at its fastest rate during the remaining 291 time period, allowing for replanting, and taking into account the age distribution of the woodland 292 resource being assessed. The worst-case loss is then the worst (highest) of these values. We 293 demonstrate an example of this in the Woodland Carbon Code case study, using possible years of 294 arrival at 5-yearly intervals. 295 296

Calculation of individual tree species risk factors 297 298
In the credit risk example in Equation (1), the risk-weighted assets for individual positions are 299 summed to give the overall risk for an entity e.g. department or bank as each exposure is mutually 300 exclusive. However, the consequences or impacts of each P&D are not mutually exclusive as they 301 affect the same woodland resource. More than one P&D could attack the same tree at the same 302 time, however, P&D could also attack successively. For simplicity we have assumed that P&D act 303 successively and that P&D can only impact on the remaining trees after the previous P&D have 304 attacked. For example, if P&D (p1) has caused a loss of 5% of a given tree species, and P&D (p2) is 305 estimated to cause a 10% loss of trees, then p2 can only affect the remaining 95% of trees. Therefore 306 p1 would cause a loss of 5% of the tree species, leaving 95% of the trees, and p2 would cause a loss of 307 10% of this remaining 95% i.e. 9.5%. The total loss would therefore be 5% + 9.5% = 14.5%. 308 309 Aggregation of the P&D risk factors for each tree species is therefore calculated by sequentially 310 applying the risk factors for each individual P&D threat rather than summing them. The process of 311 aggregating the P&D risk factors for a specific tree species is outlined in the sequential equations 312 shown in (4) below. This example aggregates all the P&D risk factors that affect a given tree species 313 (t1). It includes risk factors RiskFacp(1 to n)t1, where n = the number of P&D affecting tree species t1. The 314 variables z(1 to n) are used to denote the interim values as each P&D risk factor has been aggregated, 315 and which form an input to the next aggregation: 316 317 RiskFacp1t1 = z1 Aggregation of RiskFacp2t1 = z1 + (RiskFacp2t1 x (100% -z1)) = z2 Aggregation of RiskFacp3t1 = z2 + (RiskFacp3t1 x (100% -z2)) = z3 Aggregation of RiskFacp4t1 = z3 + (RiskFacp4t1 x (100% -z3)) = z4 …And so forth until: Aggregation of RiskFacpnt1=z(n-1) + (RiskFacpnt1 x (100% -z(n-1)) = Overall Risk Factor for tree species t1 = RiskFact1 The risk factors for all of the P&D that affect each tree species being assessed are sequentially 320 aggregated in this way to give a risk factor for each tree species. 321 322 Calculation of overall risk factor for the woodland resource 323 324 The final overall risk factor for the woodland resource being assessed is calculated by weighting the 325 risk factors for each tree species by the proportion % concentration of that tree species across the 326 woodland resource. So if there were three tree species (t1, t2 and t3) with risk factors of RiskFact1, 327 RiskFact2 and RiskFact3 and concentrations (c) in the woodland of c1, c2 and c3 (expressed as a % 328 proportion of the total woodland occupied by the tree species), then the overall risk factor for the 329 woodland resource (RiskFacWood) is calculated by: 330 against future losses. The size of this buffer must be determined at the project outset. Under the 345 current version of the Code (Forestry Commission 2014), this entails a percentage potential loss 346 assessment over the project duration for a specified list of risks, which includes P&Ds. The project 347 developer carries out an assessment for each type of risk, which must fall within the specified range. 348 The developer submits their methodology at verification (approval) by an independent assessor. 349 Table 1 shows the risks and ranges for the current version of the Code, the current amount set aside 350 for P&D ranging from 3 to 10%: 351 352 INSERT In this case study, we provide an overall assessment of risk to the projects certified under the 357 Woodland Carbon Code from a sample of pest and diseases, to demonstrate the application of our 358 approach, and how it might verify whether this range is likely to be adequate to cover future losses. 359 The results will be used to demonstrate how the outputs can be used to support species selection, 360 identification of priority P&D threats, and to support policy decisions such as whether the current 361 buffer range is likely to be adequate against future losses. 362

363
The time period (t) was defined as 100 years -as this was the most common duration (53%) of 364 Woodland Carbon Code projects at the time of assessment, as determined through an analysis of 365 project documentation. The geographical area (g), was defined as Great Britain: the area over which 366 the Woodland Carbon Code projects are distributed. It was decided to perform the assessment for 367 the main tree species in the portfolio, defined as those constituting more than 2% of the total area 368 of all projects, and for a sample of P&D. The analysis therefore required the following steps: 369 370  Calculation of the concentration of different tree species within the Woodland Carbon Code 371 project area to provide the weights for the calculation of the overall risk factor for the 372 woodland resource used in calculation Equation (54) and to identify those species to be 373 assessed, constituting over 2% of the project area; 374  Identification of the P&D threats to these tree species; 375  Choice of a sample of P&D to demonstrate the P&D methodology; 376  Calculation of ProbArr and ProbMax in Equation (2) for each P&D; 377  Development of a simple spreadsheet to calculate ProbLoss in Equation (2) (2)  i.e. less than 1 every 4 years. It is not possible to quantitatively derive an estimate for individual P&D 443 from this historic rate of arrival; however, it is clear that the historic rate of entry is low. An increase 444 in shipping and other channels of arrival such as imported saplings suggest that this rate may 445 increase in the future (Eschen et al. 2015). Default values were therefore set as 1 to 5%, which 446 conveniently corresponds to the 1 to 5 Likert Scale values from the Register. In the 100-year time 447

Calculation of ProbArr and ProbMax in Equation
horizon assumed for the Woodland Carbon Code projects, this therefore implies an expected arrival 448 of year 100 for the least likely (Likert Scale 1) and year 20 for the most likely (Likert Scale 5). All P&D 449 are therefore assumed to arrive at some point within the time horizon of the project. Whilst 450 subjective, these values are conservative and this was verified by the experts consulted. During the 451 consultation, some of these Likert values were modified from their Register original values based on 452 expert knowledge. 453 14 ProbMax is the maximum possible proportion of the geographic area (g) that each P&D could 455 impact. expressed as a %. For this case study, (g) is Great Britain and so experts were asked to 456 estimate the maximum % proportion of Britain by land area that the P&D range could expand to if a 457 host tree species was present e.g. the colder northern climate may limit some P&D that would not 458 survive in this region. 459

460
Calculation of ProbLoss in Equation (2) for each identified P&D/tree species combination 461 462 As per Equation (2), ProbLoss is defined as the worst-case loss for a specific tree species/P&D 463 combination that could occur in the next 100 years across Great Britain expressed as a proportion %.  Years to 100% offull potential range -the length of time the P&D is estimated to take to reach 491 its full range. If the possible range in 100 years is less than 100the full range% then this value is 492 estimated assuming a linear rate of spread. For example, black stain root disease (Grosmannia 493 wageneriis) is only expected to achieve a proportion of 10%0.10 of its range in 100 years. 494  Mortality by species (%) -the worst-case percentage mortality caused by the P&D across its 495 established range expressed as a proportion of trees lost. Where possible this should be 496 estimated based on the worst known loss that has been caused by the P&D being assessed to 497 date. Mortality rate has to relate to the defined range. In the black stain root disease example, 498 the soil range infected is in isolated foci and the associated % mortality proportion would relate 499 to the mortality rate for trees on infected soil. So in a hypothetical area where infected soil was 500 evenly scattered across a 100 km square covering 10% of thea proportion of 0.10 of the area the 501 range would be 100.10%. Mortality could be occur in 90%a proportion of 0.9 of trees on infected 502 soil only. However, for insects such as bark beetles, the range would be the geographic range of 503 the beetle but not all trees within the area would be infected, so the range could be 100km 504 square but the % mortality rate within that range would relate to the % proportional mortality 505 rate of tree species in the geographic range of the beetle. The combination of range and 506 mortality should therefore represent the worst-case loss across Britain. It should be noted that 507 P&D outbreaks often coincide with other natural events such as bark beetle outbreaks after a 508 storm. The worst-case mortality would cover these possibilities as it is the worst possible case. 509 Worst-case relates to the level of loss over the total British range so whilst losses may be high in 510 locations with severe wind damage the mortality % estimate will be lower at the national scale. 511 512 If a P&D is sub-lethal and does not cause mortality then ProbLoss is 0%. For endemic species that are 513 at 100their full% of their range, ProbLoss is determined from the estimated mortality rate. For newly 514 arrived P&D, the Scenario Tool is used to estimate ProbLoss for each P&D and tree species 515 combination. For P&D already present, but not yet at their full range, the Scenario Tool estimates 516 ProbLoss the loss for the spread to the remaining possible range only. The worst-case lossProbLoss is 517 then the sum of the mortality rate weighted by the current range, and the loss factor weighted by 518 the remaining range i.e. the area that the P&D will spread to during the project duration. 519 520 The Scenario tool spreadsheet, is based on a series of 20 tables. Each   the approach is providing an estimate for the whole Woodland Carbon Code, the Guideline default 541 recommendations of GYC4 were used, except for Sitka spruce where GYC6 was used (. note that the 542 GYC for Sitka spruce planted under the Woodland Carbon Code is typically lower than that planted 543 for timber alone). An initial plant density of 2500 trees per hectare and no thinning was assumed. 544

545
The Tool assumes that following loss, replacement trees are planted, and factors in replacement 546 carbon sequestration. Since it is not known which species would be planted, the Tool uses the 547 sequestration rates for generic SAB woodland as it is the only mixed woodland carbon sequestration 548 estimate and is the default in the Guidelines for many species without individual tables. 549 550 In this way, the Tool estimates the losses that would occur by Year 100 for each 5-yearly arrival 551 period. The worst-case loss ProbLoss is therefore the worst value (highest % proportional loss) of 552 these 20 values. This assessment is performed for each of the P&D and tree species combinations. 553 554 Calculation of P&D, tree species and woodland resource risk factors 555 556 Once the values required for ProbArr, ProbMax and ProbLoss were estimated in this way, individual 557 P&D risk factors for each P&D/tree species combination were calculated using Equation (2). They 558 were then aggregated into tree species risk factors using the sequential application of Equation (4)  559 and then aggregated into an overall estimated risk factor for the woodland resource of the 560 Woodland Carbon Code using Equation (5). 561 562

564
The main tree species selected for the Woodland Carbon Code case study 565 566 Table 2 shows the results of the analyses of tree species composition in the Woodland Carbon Code 567 projects from the two approaches: firstly the analysis of total hectares planted using project design 568 Commented [G5]: Too late now but why was such a low value used? Mean GYC for Sitka spruce in GB is 16! RESPONSE: This is not true for the WCC. I have just done a breakdown of the current WCC which shows that 51% of the Sitka planted is GYC6 or less. This is because the sites are primarily to establish native woodland. I cannot put this figure in as it represents an unpublished 2016 analysis of the current project portfolio which has over twice as many projects as this case study in 2014. I chose GYC6 at the time as it was nearest to the guideline recommendation of GYC4 as per the text and seemed reasonable from the PDDs I had reviewed. Additional sentence added. The results show that these two approaches produced similar results in terms of identifying the main 574 11 tree species: using the number of hectares planted instead of the number of trees is therefore 575 not unduly biased. As a result, we used the hectare analysis from the PDDs to determine the sample 576 of tree species for the case study as it covers all of the Woodland Carbon Code projects (whereas 577 Forest Carbon Limited projects only cover 40 of the 60 projects). All species constituting over 2% 578 were included in the sample. The key risk metrics and a demonstration of the calculation of individual P&D risk factors (Equation 593 2) for a sample of assessed pest and diseases is shown in Table 3.  594   595   INSERT TABLE 3  596   597 The mulberry longhorn beetle (Apriona germariiis) provides an example of where the final risk factor 598 is adjusted to account for the fact that it would never cover 100% all of Britain i.e. multiplied by the 599 maximum range of 50%0.5. Table 3 also shows how the risk factor varies significantly for P&Ds with 600 similar metrics but where one is present and one is yet to arrive. Since only a sample of P&D that threaten the tree species in the Woodland Carbon Code were 628 included in the case study, the initial results do not represent a final risk assessment for the 629 Woodland Carbon Code and it has not been possible to carry out any form of sensitivity analysis. 630 However, the results do demonstrate how the approach works and the applicability of its findings. 631 We are not aware of any other methods in the scientific literature that take such a holistic approach 632 to assessing the future risk of P&D to woodlands. 633

634
Overall risk assessment at the woodland level 635 636 The woodland resource risk factor, RiskFacWood, provides an assessment of the worst-case loss to 637 the woodland for the defined application. For the case study of certified projects under the 638 20 Woodland Carbon Code this value is 12.1%. This represents an estimate of the potential worst-case 639 loss of sequestered carbon sequestration from the P&D assessed in the sample. 640 641 RiskFacWood can be used to factor potential woodland loss in to risk management decisions. Since 642 this represents a worst-case loss as opposed to a forecast, the amount of loss factored into analyses 643 can be varied according to the risk appetite of management. The more conservative the approach to 644 risk, the greater the proportion of this risk factor used for analyses. 645

646
The case study of the Woodland Carbon Code provides an example of this application. A key risk 647 management decision application is whether the buffers set aside by forest projects to cover P&D 648 risk are likely to be adequate. Table 1 shows that the current buffer range for assessments of P&D 649 risk for the Woodland Carbon Code is 3-10%. If the buffer was inadequate and claims exceeded the 650 carbon credits in it, the carbon market would be undermined, as credits that have been sold would 651 have to be recalled to cover any shortfall in the buffer. The woodland risk factor of 12.1% suggests 652 that the buffer might not be adequate against future losses if the most conservative approach to risk 653 (i.e. use of 100% of this risk factor) is applied. 654 655 However, Table 5 also shows that 7.7% of this total comes from the threat to ash which constitutes a 656 proportion of 0.098.53% of the woodland and faces a significant mortality risk from both ash dieback 657 (Hymenoscyphus pseudoalbidus/Chalara fraxinea), which arrived in Britain in 2012, and the emerald 658 ash borer (Agrilus planipennis), which may arrive in future. Since projects are in their first few years, 659 management could decide, for example, to replant alternative species now, exclude sales of 660 sequestered carbon from ash in the short term, or reduce the amount of ash planted. Reducing the 661 proportion of ash in favour of tree species with lower risk factors would reduce RiskFacWood. 662 663 In this way, the method provides a simple summary of the potential impact of P&D threats on the 664 woodland resource and supports policy and management decisions to reduce this risk. 665 666 Tree species concentration risk 667 668 Table 5 also shows how the approach can identify concentration risk whereby a high degree of risk is 669 concentrated in a few species. The analysis shows that the top 5 species (birch, oak, ash, Scots pine, 670 Sitka spruce) account for a proportion of 0.67% of the Woodland Carbon Code, which is therefore 671 highly exposed to significant mortality to any of these species. Birch alone constitutes a proportion 672 of 0.27% of the woodland certified under the Code and has the second highest risk factor. 673 674 21 The approach can be used to analyse why this is the case for the sample assessed, by comparing the 675 P&D risk factors in the tree species risk factor calculations as demonstrated in Table 4. Comparison 676 of the tree species risk factor calculations revealed that the risk factor for birch is higher than other 677 broadleaves as, whilst they share key threats such as the citrus and longhorn beetles, birch faces an 678 additional significant threat from the birch borer. It should be noted that this is a demonstration of 679 the application as the comparison only relates to the P&D assessed. Once all P&D risk factors have 680 been assessed and aggregated this relative risk is likely to change. The approach helps identify which P&D are of most concern through a comparison of the risk factors 689 for individual P&D. In addition, it identifies the characteristics of P&D that exhibit the highest 690 potential threat for a specific application. 691

692
The Woodland Carbon Code case study revealed that the later in the project cycle a P&D affects a 693 woodland, the worse the carbon impact, as there is less time to replant and regrow lost carbon. The 694 highest risk factors are for those P&D that affect mature trees and could spread rapidly across the 695 country. The bronze birch borer (Agrilus anxius) is yet to arrive in the country but has a high risk 696 factor of 3.657% because it is estimated to take only 15 years to cover Britain, and could cause a 697 proportional mortality rate as high as 90%0.9, partly caused by a judgement that birch in the UK may 698 have a lower resistance than birch species in north America. The worst scenario is that it arrives in 699 year 80  and so ProbLoss will need to be estimated based on a shorter timescale than the case study. 734  Some P&D damage timber quality rather than inducing mortality. The oak pinhole borer 735 (Platypus cylindrus) is such an example. It stains timber and can therefore impact on timber 736 revenues but does not cause high mortality so it has little impact on carbon projects. 737  Similarly, some P&D only affect visual appearance, such as the horse chestnut scale 738 (Pulvinaria regalis). It causes foliar spots. Others may only affect the fruit, which impacts on 739 amenity and commercial value but not on carbon. 740

741
The ProbLoss calculation therefore needs to be adapted according to application and how loss is 742 defined. The approach can reveal hitherto unrecognised risks or indeed factors that reduce risk. The case 747 study revealed, that for tree species that demonstrate the slower rates of carbon sequestration, 748