Development of a combined marine and terrestrial biodiversity indicator: research

4. Overview of main biodiversity indicator types

4.1 Abundance and occupancy-based indicators

45. The main and most frequently-used biodiversity indicator types are those based on changes in either abundance or occupancy (a measure of geographic range size) usually accompanied by a measure of precision. These indicators can be used at a single population or species level but are often combined to form multi-species/taxa composite indicators. It is, however, worth noting that creating composite indicators using measures of change in both abundance and occupancy does create problems, as these measurement 'currencies' are not directly comparable and vary in a number of ways. This is discussed in more detail in section 5.1.

46. Both indicator types are often presented as a line on a graph and tend to rely on data from established long-term monitoring and recording schemes. Examples of abundance-based indicators include the:

• Indicators for terrestrial birds, seabirds and wintering waterbirds in Scotland,
• Living Planet Index,
• UK Indicator C4a. Status of UK priority species- relative abundance,
• UK Indicator C6: Insects of the wider countryside (butterflies),
• UK Indicator C8: Mammals of the wider countryside (bats),
• European Wild Bird Indicator (derived from the Pan-European Common Bird Monitoring Scheme, PECBMS).

47. Occupancy-based models form the basis of UK indicators on Priority Species (C4b) and Pollinating Insects (D1c), as well as nearly one third of species on the Living Planet Index of the Netherlands (van Strien et al. 2016). In the marine environment, trends in abundance or occupancy of certain marine species have been adopted as indicators in their own right, including abundance of grey and harbour seals, and of certain cetacean species, seabirds, and sensitive fish species (OSPAR 2017a, b, c, d).

48. Most published abundance-based metrics are composite multiple species trends. The creation of these involves two analytical steps; firstly, the production of annual population indices for each individual species (or, in some cases, a more categorical index, e.g. 'recovered / not recovered') and then combination of these indices into the grouping required to form a composite indicator.

4.1.1 Single species indices

49. There are several different analytical approaches to the creation of single-species indices. Many of the annual indices produced in the UK use the following statistical modelling technique (Eaton & Noble 2018, Boughey & Langton 2017) - Generalised linear models (GLMs), with full site and year effects, a log-link function, and Poisson error term to deal with the distribution of count data. At an international level, the species trends incorporated in the WWF Living Planet Index (LPI) are created using a different statistical framework. Generalised additive models (GAMs) are used to identify underlying trends in different population time-series, and these are in turn used to calculate average rates of change at a species level (WWF 2018).

50. The single-species indices resulting from such analyses are frequently smoothed before use in composite indicators. This process brings benefits, by removing short-term 'noise' that may be caused by sampling error or minor fluctuations due to, for example, weather effects; smoothing can provide a clearer measure of the underlying trend. The smoothing procedure can thus influence the assessments of change over particular periods as well as the confidence in these estimates.

51. A number of smoothing methods are available. For the UK bird and bat indicators a post-hoc smoothing spline equivalent to the application of statistical models (GAMs) is used (Eaton & Noble 2018, Boughey & Langton 2017). GAMs are a non-parametric technique in which the population trend can be set for any degree of smoothing by altering the degrees of freedom (d.f.) used in the calculation of the model. If the d.f.'s are set to one, a model in which abundance follows a linear function of time is produced (i.e. a straight line), whilst if the d.f.'s are set to equal the number of years in the time series, a model is created in which the estimates for consecutive years are simply joined (equivalent to no smoothing) (Eaton & Noble 2018).

52. Abundance indices can then be generated for each species, indexed to a value of 100 (1 is sometimes used - both options are appropriate for ease of comprehension) in a baseline year. These indices report relative changes in abundance: a rise from 100 to 200 reflects a doubling in numbers, a decline to 50 a halving, relative to numbers in the baseline year (Eaton & Noble 2018). Species indices are often presented in both smoothed and unsmoothed forms; however, assessments of change over set periods are usually based on the smoothed version. Smoothing does, however, mean that the estimates for the final year of a trend must be treated with caution as they lack the smoothing effect of data in subsequent years. The nature of smoothed trends, in that data from any given year has an impact on trend values for earlier (and later) years means that existing species indices (and hence indicator) values will be different in subsequent annual revisions (Eaton & Noble 2018). Unsmoothed indices are often used to measure change over the final year of a trend sequence.

53. Confidence limits around species trends are usually generated by bootstrapping; i.e. repeated resampling (with replacement) to generate a sample of estimated trend values, with the 2.5% and 97.5% percentiles giving the 95% confidence limits around the trend value for each year (Efron 1982). Some data sources do not allow the calculation of error in the trend estimate due to the structure of the data collected (e.g. from non-randomly survey sites, and sites of differing size); for instance, Wetland Bird Survey (WeBS) and Seabird Monitoring Programme.

54. Depending on the purpose of individual indices, alternatives to these continuous trend approaches may be used. For instance, in demersal fish species, an aim under the MSFD has been to determine whether species are recovering, or are not undergoing further decline, as a response to management efforts. To this end, the whole time series of a survey has been used as the reference period, and a species is classified for an assessment year based on whether its abundance falls in the top 25th percentile of all recorded abundances (recovery), or above the bottom 25th percentile (no further decline) (Greenstreet et al. 2012). This shows that similar data have been processed in contrasting ways depending on specific policy and management objectives.

55. Recently, species indices have been created from presence-only biological records. Species trends from such data are robust if analysed appropriately, for example using Bayesian occupancy-detection models (Isaac et al. 2014). The application of this approach in the UK has used data from national recording schemes held by the Biological Records Centre (BRC); these records are verified by species experts. Datasets are compiled into species assemblages, such as all species of bee, and records from the same date and 1-km² square (records at a coarser spatial scale are excluded from analysis) are considered to constitute 'site visits' and the number of species for each site visit calculated as 'list length' – a measure of effort. A Bayesian occupancy-detection model is then fitted for each species, using two hierarchically coupled sub-models: one models occupancy (i.e. presence versus absence of each site-year combination), and the other models detection (i.e. recorded versus not-recorded on each visit). Since true occupancy is unknown, this form of occupancy-detection model is of a class of statistical models known as 'hidden process' or 'state space' models. The 'list length', defined above, is used as an estimate of sampling effort. The 'season' (also known as the closure period) in these models is the year (i.e. occupancy for each year, using all the recording visits that took place during that year, is estimated). Recent implementations of occupancy-detection models have enabled the approach to be used for species with sparser datasets than previously, meaning that occupancy trends can be produced for more species in more taxonomic groups, and reduces problems of under-representation of rarer species in the data. This development uses prior distributions in a Bayesian framework, which allow the probability of occurrence at a site in a given year to inform the probability of occurrence at that site in subsequent years, in a biologically plausible manner and can produce trends even where available data are sparse (Outhwaite et al. 2018).

4.1.2 Multi-species composite indicators

56. Once single species indices are created, multi-species indicators can then be calculated. The approach used for UK birds, bats and butterflies is outlined in Gregory et al. (1999) and use the geometric mean of the constituent species indices. Using the geometric mean means that a doubling in the population index of one species is balanced by a halving of another (Buckland et al. 2005). Methods for the creation of such indicators are discussed in more detail in section 5.1.

57. As with the species indices, some existing biodiversity indicators exist in both smoothed and unsmoothed forms. Where the species data are already smoothed (e.g. using a GAM), the gains from additional smoothing procedure may be limited. An alternative is to smooth the headline indicator, rather than the constituent species indices. The Multi-Species Indicator (MSI) tool developed by Statistics Netherlands (Soldaat et al. 2017) does this: it is used to create supranational indicators such as those produced by the Pan-European Common Bird Monitoring Scheme (e.g. Klvanova et al. 2009). A similar method has recently been developed at the Centre for Ecology & Hydrology (Freeman et al. 2020; see paragraph 139 for further detail), with the specific purpose of creating smoothed headline indicators from diverse datasets containing missing values. Both the MSI and the Freeman method are implemented in a Bayesian framework and make it possible to account for uncertainty in the species index values.

58. Marine indicators have again developed in a related but somewhat different way, reflecting the trends-based targets set by the MSFD and other drivers. For instance, in the demersal fish index individual species have been classified by their recovery status, and then this is summarised across species as simply the number (or proportion) of species meeting the relevant target (Greenstreet et al. 2012).

59. Composite indicators tend to give equal weighting across the species included. However, the use of weighting can bring benefits, such as addressing any biases within the availability of species trends relevant to the indicator. For example, we know that despite great strides in the availability of biodiversity data within Scotland, the trends available for indicator construction will not be representative of terrestrial and marine biodiversity as a whole. Potential sources of bias that might be addressed through the use of weighting include:

• Taxonomic bias – for instance, trend data will be available for more vertebrates than invertebrates, and within invertebrates there are biases towards groups such as Lepidoptera (butterflies and moths). Burns et al. (2018) investigated the impact of weighting upon indicator outputs and found the impact of controlling for taxonomic bias through weighting was sensitive to the taxonomic level (i.e. phylum or kingdom) at which weighting was deployed.
• Habitat – due to issues such as ease of access, there are disparities between the volumes of data available for species within different habitats. This presents two issues, in that trends for individual species may be biased towards certain habitats (although this can be addressed within trend analyses), and the species composition within indicators may be biased towards those using widespread and accessible habitats (e.g. montane species may be underrepresented). This could be particularly relevant to combining terrestrial and marine elements of a combined indicator, with robust trend data likely to be available for a substantially higher proportion of terrestrial species than marine.
• Ecological function – biases might favour higher trophic levels (due to the wider interest and hence the availability of data on vertebrates), and disfavour other functional groups such as detritivores for which data are scarce or absent. Such biases can be difficult to control due to the lack of systematic data on such traits.
• Conservation status – there may be a bias in trend availability towards species of higher conservation concern such as those on the Scottish Biodiversity List (both because species are only likely to be designated as being of concern if sufficient monitoring data are available, and because designation may subsequent increase the likelihood of monitoring). Whilst species of concern may be faring more poorly than those not, designation as a conservation priority means a species is more likely to become the recipient of targeted conservation action.
• Commercial importance - in marine systems, species that are commercially exploited in fisheries are often surveyed more systematically than unfished species.
• Abundance – sample sizes, which influence the ability to produce trends, may be larger for widespread and abundant species leading to a bias.

60. Even apparently simple issues, such as addressing taxonomic biases in data availability, can present difficult decisions. Weighting can be used to correct biases in the availability of species trends for example to upweight the contribution of under-sampled taxonomic groups, although it cannot correct for biases in instances where no data is available at all. The impact of such an approach will vary, however, depending at what taxonomic level (e.g. family, order) weighting is conducted (Burns et al. 2018). Such an approach will also perpetuate imbalances caused by some taxonomic groups being more biodiverse than others. If, for example, we had data for all of Scotland's naturally occurring species, the impact of changes in vertebrate population upon a composite indicator would be minute, unless a weighting approach was employed to control for the greater diversity within some taxonomic groups. Conversely, we may need to address how to incorporate data reported at an amalgamated level – for instance, trends from long-term monitoring of zooplankton, an important measure of marine ecosystem health, are not available at the species level.

61. The best available data sources for some species do not allow indices to be produced for the complete time period required for a composite indicator; in such cases, provision must be made to allow species to drop into and out of the indicator according to data availability. If the indicator is set to a baseline year before the entry date of such species (e.g. if it is baselined to its start year) then they can be entered into the composite indicator at the mean value of the indicator for the year in which they enter. This ensures that the addition of new species does not have an artificial impact upon the composite indicator (Noble et al. 2004). Similarly, protocols are available to deal with species indices that do not run to the final year of an indicator (Eaton & Noble 2018).

62. Composite indicators, an aggregation of individual species indices showing average trends, can hide a large disparity in the fortunes of the constituent species. Increases and decreases in individual species can balance each other out, leading to a relatively stable index over time. However, additional supporting information, such as categorical change and ratio values outlined by Eaton & Noble (2018), can be produced to inform on this in addition to the publication of the disaggregated data.

63. Indicators based on Bayesian occupancy-detection models follow similar procedures. The headline statistic measures changes in the geometric mean occupancy across species. Uncertainty around the headline value is straightforward to calculate in a Bayesian context, since the species indices are presented as a distribution of estimates, rather than a point. Moreover, the relationship between data and model permits an intuitive interpretation of the uncertainty. For example, if 95% of the credible estimates are in one direction (e.g. a decline) then we can be 95% confident that species in the indicator have declined, on average.

64. In the creation of any indicator based upon multiple species trends, proper consideration must be given to ensuring constituent trends are of sufficient quality as to ensure the robustness of the resulting indicator. The UK's existing biodiversity indicators rely mainly upon data that originates from well-established long-term monitoring schemes (e.g. the UK Breeding Bird Survey and the UK Butterfly Monitoring Scheme). Such schemes employ rigorous stratified random sampling design, strict standardised protocols around survey methods and use quality assurance procedures relating to data collection, data collation, verification, storage, trend analyses and composite indicator construction. This allow statistical corrections to be applied to counter spatial and other biases. Even where schemes incorporate a non-random sampling approach (e.g. due to the aggregated nature of the species being monitored such as in the Wetland Bird Survey and the Seabird Monitoring Programme) temporal changes in population abundance can still be estimated by repeat coverage of the same sites (Eaton & Noble 2018). Even so, the trends derived from such data sources need to be screened for suitability, although approaches used to do so differ between indicators, variously using factors such as timescale, frequency of update, sample size, precision and concerns over bias to screen species data.

65. Information on species abundance, such as that collected by many of our long-established biodiversity monitoring schemes, represents the highest quality data for the creation of biodiversity indicators; however, in practice, they cover a relatively small proportion of the total number of species found in the UK. Many taxonomic groups, particularly invertebrates, are poorly represented. Opportunistic biological records, collected in relatively unstructured ways, are increasing being used to explore trends in a greater range of species (e.g. Hayhow et al. 2016, 2019). Statistical methods to account for biases caused by sampling effort, spatial coverage and detectability, are being developed to provide robust estimates of occupancy over time (Isaac et al. 2014, Outhwaite et al. 2018). Thus, the quality of the data on species indices is a function of both the availability of raw data and the statistical techniques used to analyse them.

4.2 Red List Indices

66. The IUCN Red List is widely accepted as a robust system for assessing the conservation status of species, specifically with regards to their risk of extinction. A standardised set of criteria, using quantitative thresholds based on population size, population trends and area of distribution, enable species to be assigned to a category based on relative risk of extinction. These categories range from Least Concern, through Near Threatened and then three categories of threat (Vulnerable, Endangered, Critically Endangered) as well as Extinct, Extinct in the Wild and Data Deficient (IUCN 2012a).

67. Whilst other systems of assessing conservation status (e.g. Birds of Conservation Concern for birds in the UK, Eaton et al. 2015a) exist, none have the universal applicability of the IUCN Red List. Whilst there are challenges in ensuring the uniform applicability of the IUCN assessment process to different taxonomic groups, differing levels of data availability, and different spatial scales, the system is designed to enable application in all circumstances. While originally developed for use at a global scale, the development and subsequent refinement (IUCN 2012b) of guidelines for the application of the Red Listing process at a regional scale has resulted in a proliferation of Red Lists at continental (e.g. for birds in Europe, BirdLife International 2015) and national (e.g. Stanbury et al. 2017a) scales; this regional process is designed to be used at any spatial scale although concerns have been raised about the validity of the process at smaller spatial scales (e.g. Charra & Sarasa 2018). The regional assessment process is two-stage, with the first stage applying the global assessment process, followed by a second stage considering the impact of populations of the same taxon found outside the region of interest, e.g. whether such populations offer the possibility of a 'rescue effect' and so reduce the extinction likelihood within the region itself. Within the UK, regional IUCN assessments are generally applied for Great Britain, and/or all-Ireland, as these are more appropriate biogeographic units than the political area of the UK. At present approximately eight thousand species (Conservation Designations for UK taxa – collation) have IUCN Red List assessments for Great Britain, although as yet none of these have been assessed more than once using the modern regional Red List process. No assessments have been conducted at a Scotland-only scale, although for a small number of species that only occur in Scotland within Britain, British-scale assessments are de facto Scottish assessments.

68. The Red List Index (RLI) uses available IUCN Red List assessments to measure overall trends in extinction risk for given sets of species and geographical areas, based on the number of species in the different categories of extinction risk (Butchart et al. 2007). Initially developed for use at the global scale, the RLI can also be calculated for any spatial scale that assessments are available, and for any given taxonomic group or combination thereof.

69. The RLI is calculated by multiplying the number of species in each Red List category by a category weight, summing the products of this across all categories, and then expressing this as the proportion of the maximum possible product, whereby a RLI value of 1.0 equates to all species being of Least Concern. The lower a RLI value falls below 1.0, the greater the level of extinction risk across the species included in the index. In order to be of value as a measure of change in extinction risk through time, the RLI requires species to be reassessed for extinction risk at intervals – it can be calculated for any group of species that have been assessed at least once. It is important that the RLI reflects genuine change, either in the form of improvement or deterioration in extinction risk; however, changes in assessed risk are frequently due to revised knowledge, and these need to be excluded from the calculation.

70. At present global RLIs are published for birds, mammals, amphibians, reef-forming corals, and cycads. Given the uneven taxonomic spread no attempt has been made to combine these into an overarching indicator. An initiative to broaden the taxonomic spread of repeated assessments by sampling species from a wide range of taxonomic groups, such as dragonflies, fish, dung beetles and monocotyledonous plants (grasses, lilies, orchids etc.) is intended to enable the production of a more representative Sampled Red List Index (SRLI), although the relative scarcity of Red List assessments of marine species (Webb & Mindel 2015) may limit use of the SRLI as a fully integrated biodiversity indicator.

71. The RLI approach has been widely accepted as suitable for measuring and reporting changes in biodiversity status at a high level, and is particularly suited for assessing progress towards target 12 of the Aichi 2020 targets, which states "By 2020, the extinction of known threatened species has been prevented and their conservation status, particularly of those most in decline, has been improved and sustained". Global RLI are used for reporting progress towards this target (Secretariat of the Convention on Biological Diversity 2014) as well as progress at a national level, aided by specific guidance (Bubb et al. 2009). The use of Red List criteria which are designed to be used even in the absence of robust species data (e.g. abundance trends) means the RLI can be calculated for geographical areas and taxonomic groups for which such data are scarce, given sufficient time for repeat assessments to be made. There are, however, shortcomings, with RLI being relatively insensitive to change (given the coarse level of resolution – species can undergo very considerable status change without changing Red List category), and having low temporal resolution, given the often long periods between species being assessed.

4.3 Diversity metrics

72. The role of anthropogenic pressures in perturbing the dynamics between different species within communities has long been recognised, and so 'diversity' metrics are well-established to measure changes in the relative abundance of species. Indices such as Shannon, modified Shannon and Simpson's are used to measure changes in community structure, often on local sites (e.g. for specific communities).

73. Simpson's index (Simpson 1949) is a diversity metric which accounts for both the number of species and their abundance. If d_ij is the number of individuals in the system (abundance) in year j that belong to species i and p_ij = d_ij/∑_id_ij the proportion of them from all species. Simpson's index is then Dj = ∑_ip²_ij. The transformed index – log_e D_j is used as a diversity metric, with low values indicating dominance of a few species, high values meaning higher evenness of population sizes. The Shannon index (Shannon 1948) is a similar diversity metric, again with low values when a few species dominate and high values when none do. It is defined as H_j = −∑_ip_ij log_e(p_ij). If, however, all species increase and decrease with a similar rate (and thus any unevenness remains the same), both Simpson's and Shannon's indices would remain unchanged – not ideal if they are intended for use as indicators of changes in biodiversity. Buckland et al. (2005) proposed a modified Shannon index to address this, whereby abundances of species in all years are divided by the summed abundances of all species in year one.

74. Such measures are typically used to make spatial comparisons, rather than measure trends through time, but diversity indicators could be of considerable use in measuring the impact of drivers of biodiversity change such as landscape-scale degradation, which can result in generalist species increasing whilst specialists (often in habitat use, although niche breadth can be defined in other ways) decrease. Buckland et al. (2017) reviewed approaches to measure diversity change over time and proposed that measures that consider turnover between species (e.g. Yuan et al. 2016), rather than simply diversity, might be most appropriate for measuring changes arising from the impact of anthropogenic drivers.

75. Various measures of diversity and community composition have been adopted as indicators in marine systems. For instance, Rombouts et al. (2019) present a test of measures of phytoplankton community change, incorporating both changes in alpha diversity as well as temporal change in community composition. In marine fish communities, size structure is often considered a useful indicator of community composition, with indices such as the Large Fish Indicator (proportion of individual fish that are over some nominal size, e.g. 30 cm) being widely employed as a simple measure of how fish community structure and diversity changes through time (e.g. Greenstreet et al. 2010, Mindel et al. 2018).

4.4 Biodiversity Intactness Index

76. The Biodiversity Intactness Index (BII) was first proposed in 2005 (Scholes & Biggs 2005) as an attempt to quantify loss of biodiversity compared to 'intact' pre-modern abundance levels, by measuring change at the site rather than species level. This is done by combining estimates of local abundance with data on land use and other anthropogenic pressures in order to model likely abundance levels, and compare these to predicted intact abundance, using fine-scale (1-km) remote-sensed datasets of these pressures. Subsequent development and massive data collation have enabled the production of global indicators of BII (Newbold et al. 2016), with further refinements of methods (Purvis et al. 2018). The method is scalable, meaning that indicators can be produced at any spatial scale given sufficient data; for example, estimates for BII in the UK's four countries were presented in Hayhow et al. (2016). At the global scale the BII has been adopted at a high level, including by the Intergovernmental Platform on Biodiversity and Ecosystem Services as an indicator of progress towards CBD Aichi targets 12 and 14. However, there is considerable debate about the robustness of the BII, with concerns about its precision at smaller spatial scales, and evidence that it substantially underestimates loss compared to intact levels (Martin et al. 2019).

4.5 Essential Biodiversity Variables

77. Essential Biodiversity Variables (EBVs) are a recent concept (Pereira et al. 2013), modelled on the existing Essential Climate Variables, and are intended to serve as the minimum set of parameters required to be measured for the robust monitoring of biodiversity status at a national scale, and by amalgamation at a global scale. If such a suite of measures could be identified and agreed upon at a global scale they would serve as the basic units with which to study, report and manage biodiversity change, and thus would inform the development of global indicators and enable a harmonised monitoring system to be developed.

78. At present, six classes of EBV have been identified – genetic composition, species populations, species traits, community composition, ecosystem function and ecosystem structure. Between two and six potential EBVs have been suggested for each of these classes. For example, the species populations class includes the suggested EBVs of species distribution, species abundance, and population structure by age/size class.

79. An ideal EBV should be: able to capture critical scales and dimensions of biodiversity; biological; a state variable; sensitive to change; ecosystem agnostic; technically feasible; economically viable; and sustainable in time (GEO BON 2017).

80. EBV's are themselves not indicators, but could be regarded as the building blocks of indicators, and a common currency on which to base biodiversity indicators. Whilst an interesting concept for structuring the requirements for the monitoring and reporting of biodiversity change, further development is required before EBV's offer a practical approach to reporting at an overarching level.

4.6 Non-species metrics

81. It could be argued that there is no place for non-species-based metrics in the development of a combined biodiversity indicator, as an indicator measuring change in biodiversity should be based on the status (e.g. population trends) of species.

82. Measurements of biodiversity can be made at more than one level, with genes, species, and ecosystems being the most typically employed scales. The indicator approaches reviewed so far are those that consider species' status as the basis for measuring change in biodiversity, and this is the commonest approach in usage, and likely to be most appropriate for the purpose of a biodiversity indicator for Scotland. However, biodiversity varies below species level, in the genomes and genotypes of individuals and populations, and can be considered at the broader scale of ecosystems; communities of individuals of multiple species that coexist and interact within a given area.

83. Indicators at genetic and ecosystem scales are not as well developed as those for species, and have not been adopted as high-level measures of biodiversity status. Many questions remain: for example, there is no consensus over how best to measure genetic diversity e.g. whether to measure diversity in genotypes, or in genomes. At present most developments have focussed on indicators of commercially-valuable genetic variation in livestock or crop plants, such as indicator C9a of the UK Biodiversity Indicators, which describes trends in rare and native breeds of farm animals.

84. Despite the greater suitability of species-based metrics, there are circumstances where such non-species information could be useful. Non-species metrics may:

• act as proxy measures for biodiversity, if these are appropriately backed by studies that relate biodiversity impacts to the measured metric. For instance, measures of habitat areas subject to nitrogen deposition above their critical loads provides information regarding both the impacts of pollution on species and on ecosystem processes. There is, however, less information available concerning lag times for recovery, so current deposition levels may not be reflective of current biodiversity trends. Habitat area in itself can be a useful metric, though there is an inherent problem in relating area to habitat quality and hence biodiversity.
• partially counter the taxonomic bias of some aggregate metrics if backed by suitable studies showing the impact of what has been measured across taxonomic groups; for example, measures of the status of water bodies is relevant for all species within the system. However, an integrative measure cannot account for individual drivers having impacts on specific species or species groups.
• provide information concerning the balance between target/protected species and species that are out of place or invasive. For instance, habitat condition assessments generally provide information regarding both desirable and undesirable species. However, they are focussed at common species as rare species appear rarely in the habitat level sampling carried out.

85. Potential non-species metrics fall into a range of categories:

• Area of habitat - Potential metrics include the extent of well-defined habitats, area under agri-environment management, certified forest, grass cut for hay, fallow/set-aside, High Nature Value Farming, (inverse of) soil sealing and the area of statutory designated sites. All show correlation with biodiversity value, as areas under these types of management should have higher levels of biodiversity than similar areas managed more intensively. Correlations may, however, be weak; for example, areas cut for hay may have previously lost their characteristic hay meadow vegetation and invertebrates, or the evidence for this correlation may be lacking - Scottish data on the success of agri-environment is lacking other than for a few species (e.g. Corn Bunting, Perkins et al. 2011). Habitat area is the main driver of population size for many species (Fahrig et al. 2019) but connection/fragmentation is also important (Horváth et al. 2019), so a measure of habitat fragmentation would be a useful indicator of the ability of species to move through landscapes. However, all area metrics are dependent on high quality, repeatable data sources available at appropriate habitat resolution; products based on data such as the land cover map of Great Britain could be developed but are not currently available. In marine benthic systems, habitat-focused indicators have formed the basis of most assessments and are integral to both OSPAR (Convention for the Protection of the Marine Environment of the North-East Atlantic) and MSFD, with targets related to the distribution, extent, and condition of various key sediment, rocky, and biogenic habitats (OSPAR 2017, Defra 2014).
• Site condition monitoring has been a feature of designated sites (Sites of Special Scientific Interest) for many years. Initially sites were visited on a six-year cycle but recently visits have been based on a risk assessment of likely changes: for example, geological sites designated for solid geology were generally deemed low risk and visited less frequently. Site condition provides an overall assessment of the quality of the habitat assessed against a template of what a good example of that habitat should look like. However, the method has rarely been assessed in terms of how well it captures the biodiversity value of a site (MacDonald 2003) and it suffers from repeatability problems (MacDonald 2010). The proportion of natural features of designated sites in favourable condition (a combination of the three categories 'favourable', 'unfavourable – recovering' and 'unfavourable – recovering due to management change') is reported annually by SNH in an official statistics publication.
• Water quality - A range of water quality indicators are routinely collected from many marine and freshwater bodies, including measures of nutrients, such as phosphate and nitrate; elevated levels of which have been correlated with reduced biodiversity (e.g. Lambert & Davy 2011). Similarly, high levels of water abstraction can also have substantial impacts on biodiversity (e.g. Flavio et al. 2017) as can poor coastal bathing water quality. More integrative is SEPA's condition assessment of water bodies (Scotland's environment 2019), which uses data on water quality problems arising from discharges of pollution and diffuse pollution running into rivers from agricultural land and urban areas, modified river flows and river channels, barriers to fish passage and the presence of aquatic invasive non-native species.
• Agricultural intensification is known to have negative impacts on biodiversity (Tscharntke et al. 2005). Fertiliser and pesticide use are negatively correlated with biodiversity, including a strong relationship between eutrophication and plant species richness (e.g. Firbank et al. 2007) and pesticide use and bumblebee colony productivity (Goulson et al. 2015). However, national or regional figures may not provide useful data for an indicator because, for example, an increase at a large spatial scale may either indicate low-level, widespread increases or substantial, localised increases.
• Air pollution is an important driver of biodiversity, with demonstrable effects on biodiversity in a range of habitats (e.g. Bobbink et al. 1998). Experimental and survey research has developed a range of critical loads; the amounts of pollution deposition above which impacts are detectable. Combining habitat and deposition maps provide an assessment of the area of habitat experiencing pollution above the level which impacts plant communities. However, as levels of pollution are currently declining, there is less knowledge of how quickly habitats recover and what potential lag times might occur before recovery is seen.
• Invasive species are established as a driver of biodiversity loss (e.g. Hooper et al. 2012), but there is little data on the impacts of most invasive species on biodiversity, except for species such rodents (Stanbury et al. 2017b), mink (Moore et al. 2003), rhododendron (Hulme et al. 2015) and American signal crayfish (Holdich et al. 2014), and, in the marine environment, species such as the Australian tubeworm and the bay barnacle (Katsanevakis et al. 2014) Additionally, invasive species include novel diseases such as ash dieback (Mitchell et al. 2014), as well as vectors of disease (e.g. the invasive crab Rhithropanopeus harrisii which has spread white spot syndrome to commercially harvested shrimp; Katsanevakis et al. 2014) where impacts may take years to develop after the first evidence of the disease. The inclusion of species, such as invasive species, as a negative contributor to a biodiversity index is attractive as it moves beyond treating all species as equivalent. Note that non-native species are excluded from most species-based indicators including existing wild bird indicators in Scotland and the UK, as trends in these species are felt inappropriate for an assessment of the status of biodiversity and, by inference, the wider environment.

86. Non-species metrics may have a role to play in the development of a biodiversity indicator, but their use in such an indicator is hindered by a number of characteristics. For example, substantially different data units are used across the various measures and changing methodologies over time may make it difficult to use some metrics, including those describing habitat condition. There may be overlap with other indicators used in the National Performance Framework, such as the Natural Capital Asset Index, so useful indicators such as those relating to water and air pollution may already be represented within the NPF suite.

87. Moreover, such indicators are as best regarded as proxies for what changes might be happening in biodiversity. Changes in such non-species metrics might reflect changes in biodiversity as both are subject to the same drivers and might respond in the same way (for example, if the condition of protected sites deteriorates, we might expect biodiversity to also decline), or the non-species metric might be the driver of change itself (e.g. air pollution). But the relationship is rarely close: species may have a large proportion of their populations outside of designated sites, for example, or be influenced by many factors in addition to air pollution. There are exceptions, for example tropical deforestation can be taken as a good measure of biodiversity loss given the scale of the impact it has, and in the absence of robust data on biodiversity trends, such proxies would be better than nothing in informing what might be happening. In Scotland, however, there are robust measures of changes in species dating back over a number of decades which mean that resorting to the use of non-species proxies should not be necessary.

4.7 Summary

88. We have described the main types of biodiversity indicators in use currently, in the UK and more widely. Most of these are derived directly from data on species, measuring change in either abundance or occupancy (distribution).

89. There are a number of ways this species' trend data can be used to develop biodiversity indicators. Although single species' trends can be used to report on change in nature more widely, the commonest approach is to combine species' trends into multi-species indicators. We have discussed the methods used to do this, and technical issues that arise, for example weighting to address biases in data availability.

90. Species' data can be manipulated in other ways to produce biodiversity indicators. These include: Red List Indices, based on average threat of extinction; diversity indices which measure changes in relative abundance between species which can reflect changes in community composition; and Biodiversity Intactness Index, which measures loss from a hypothetical intact state.

91. We also consider a range of measures based on data other than on biodiversity itself, which might inform about pressures acting on biodiversity, or act a proxy for changes in biodiversity should direct biodiversity data be missing. There are a wide range of such measures, such as habitat extent and condition, pollution, and populations of non-native species. There is not, however, a compelling case for their use given the availability of robust species' trend data for Scotland.