Pollinator monitoring more than pays for itself

of 30% pollination by high-quality scientific data, monitoring schemes at least £1.5 on data collection per £1 spent. Our findings demonstrate that long-term systematic monitoring can be a cost-effective tool for both answering key research questions and setting action points for policymakers. Careful consid-eration must be given to scheme design, the logistics of national-scale implementation and resulting data quality when selecting the most appropriate combination of surveyors, methods and site networks to deliver a successful scheme.


| INTRODUC TI ON
The abundance and diversity of pollinating insects, such as bees and flies, is critical to ecosystem functioning, crop productivity, farm income and access to nutritious food (Garibaldi et al., 2020;Genung et al., 2017;IPBES, 2016). Concerns over pollinator declines have resulted in major national (e.g. DEFRA, 2015) and international (e.g. CBD, 2016;IPBES, 2016) policy demands for information on pollinator status and trends to develop appropriate conservation and management strategies.
Recent research suggests that the occupancy of bee and hoverfly species has declined by an average of 25% across Britain since 1980, particularly among specialist species (Powney et al., 2019), resulting in homogeneous pollinator communities (Carvalheiro et al., 2013) and disrupting plant-pollinator networks (Redhead et al., 2018). These and other key studies of pollinator trends (e.g. Kerr et al., 2015) are based solely upon species records collected opportunistically by volunteer recorders rather than through repeated, standardised surveys. Consequently, it is not possible to reliably estimate changes in pollinator abundance (a major driver of pollination services; Garibaldi et al., 2020) at a local or national scale, or to identify areas of pollination service deficit (Garibaldi et al., 2020;. While drivers of pollinator decline (e.g. land use, climate change and pesticide exposure) have been inferred through field research and statistical modelling of opportunistic records (Kerr et al., 2015;Senapathi et al., 2017;Sponsler et al., 2019), without long-term abundance data we cannot reliably estimate their relative importance in driving declines at multiple scales. Furthermore, in the United Kingdom (and elsewhere), management for wild pollinators is only partially targeted, mostly comprising of agri-environment subsidies paid to farmers for implementing agri-environment measures, such as maintaining hedgerows and flower-rich field margins (e.g. DEFRA, 2020). Although these measures have been observed to effectively increase pollinator activity, diversity and pollination services at local scales (Garratt, Senepathi, Coston, Mortimer, & Potts, 2017;Scheper et al., 2013), this more likely reflects shortterm 'sinks' for flower visitors rather than true population changes (Scheper et al., 2013), with real population changes taking years to fully establish (Blaauw and Isaacs, 2014; but see Carvell et al., 2017;Morandin, Long, & Kremen, 2016).
Understanding how land management affects pollinator abundance and diversity in combination with other drivers is necessary to design more targeted, adaptive management strategies at national scales (Garibaldi et al., 2020;Lyons et al., 2008). To this end, systematic monitoring of pollinators and pollination services has been identified in the United Kingdom (DEFRA, 2015) and internationally (IPBES, 2016), as vital to obtaining a more complete picture of pollinator status and trends, identifying the importance of different pressures, and to inform suitable response measures.
Approaches to biodiversity monitoring are diverse . Large-scale and long-term surveillance monitoring (e.g. reporting of species occurrence by volunteers) provides broad, spatiotemporal baseline data that can allow early detection of issues and assessment of species trends over time (Lindenmayer et al., 2013), but that can be difficult to integrate into specific conservation management due to the nature of the data it provides (Nicholas & Williams, 2006). In contrast, targeted monitoring with a specific question or focus may be more efficient at addressing specific management issues and developing adaptive management responses (Nicholas & Williams, 2006), but may be less effective at establishing baseline data or discovering 'unknown unknowns' (Wintle et al., 2010). These potential trade-offs value of UK crop yield lost following a 30% decline in pollinators was estimated at ~£188 M/year.

Synthesis and applications.
We evaluated the full costs of running pollinator monitoring schemes against the economic benefits to research and society they provide. The annual costs of monitoring are <0.02% of the economic value of pollination services that would be lost after a 30% decline in pollination services.
Furthermore, by providing high-quality scientific data, monitoring schemes would save at least £1.5 on data collection per £1 spent. Our findings demonstrate that long-term systematic monitoring can be a cost-effective tool for both answering key research questions and setting action points for policymakers. Careful consideration must be given to scheme design, the logistics of national-scale implementation and resulting data quality when selecting the most appropriate combination of surveyors, methods and site networks to deliver a successful scheme.

K E Y W O R D S
biodiversity monitoring, biodiversity policy, cost-benefit analysis, ecological economics, pollination services, pollinators, power analysis, science policy mean there is growing interest in the relative costs and benefits of different monitoring approaches and their outcomes (McDonald-Madden et al., 2010;Nicholas & Williams, 2006). Integrating research and monitoring could potentially provide baseline information for more targeted conservation management and save costs on essential discovery science. There is evidence to support the potential for active hypothesis testing through careful, stratified sampling of a sufficiently large network (Staley et al., 2016).
As data on the state of wild pollinator populations at national scales are limited, well-designed monitoring will have inherent value in providing the consistent baseline data necessary to transition towards more targeted assessments or specific management decisions (McDonald-Madden et al., 2010;Nygard et al., 2016).
However, only Lebuhn et al. (2013) to date have explored the costs of a dedicated pollinator monitoring network, finding that a global network of 200 sites, each sampled fortnightly for 2 years within a 5-year window would cost $1.7 M. However, this study did not place these costs in the broader context of the benefits of pollinator monitoring, or the conservation of pollination services which add $235-577 bn/year to global agricultural productivity (IPBES, 2016), several orders of magnitude greater than the scheme costs.
Biodiversity monitoring is often conducted through citizen science schemes, where volunteer members of the public collect data following a standardised protocol (e.g. the UK Butterfly Monitoring Scheme; Roy et al., 2015). Although citizen science projects have returned scientifically valuable data on pollinators (e.g. Le Feon et al., 2018) and pollination services (Birkin & Goulson, 2015), significant expertise and microscopic examination are often required to identify many pollinator species. Consequently, observation-based volunteer data are often less accurate Roy et al., 2016), or of lower taxonomic resolution (e.g. Mason & Arathi, 2019) than would be collected by professional staff with training in invertebrate taxonomy (but see Ratneiks et al., 2016). By contrast, professionally led monitoring can take advantage of existing capacity to return higher-quality data in a much shorter timespan, but is more expensive .
Here, a partnership of stakeholders developed four potential, monitoring schemes, designed to identify national-scale trends in the abundance of insect pollinators (at different levels of taxonomic resolution) and/or pollination services to crops. These schemes are as follows: (a) professional pollinator monitoring, (b) professional pollination service monitoring, (c) volunteer collected pan traps and (d) volunteer focal floral observations. They represent a spectrum in data quality and annual financial investment, combining different sampling methods and levels of participant expertise to deliver defined outputs (Supporting Information Annex 1). Estimated scheme costs, based on existing pilot work, are contrasted against the economic benefits of each network (a) for research funders, by designing the network to address scientific questions and (b) for society, by offsetting the risks from failing to respond to pollinator declines that impact on biodiversity, crop pollination services and human well-being.

| Scheme designs
Four hypothetical national-scale monitoring schemes were developed, based on combinations of participants (whether data were collected and/or identified by volunteers or professionals, non-experts or experts), methods (how pollinators or crop pollination were sampled and samples processed) and metrics (what data were generated e.g. species abundance) with a focus on wild bees and hoverflies as key insect pollinator groups. Each 'Recorder-Sampling Method-Metric' combination was given a score based on feasibility and the degree of training required. This exercise considered 15 different methods, related research and using expert opinion across the partnership of authors O'Connor et al., 2019)   which has been demonstrated to accurately measure species identity from specimens collected in pan traps (Creedy et al., 2019).

| Power analysis
To develop credible scheme structures for national-scale assessments, statistical power analyses were conducted using available datasets from systematic surveys measuring pollinators and pollination services to UK crops and simulating a range of potential scenarios of change over a 10-year period. This aimed to estimate the minimum levels of replication (number of sites) required to achieve power greater than 80% to detect national-scale changes of a given size in the relevant output metrics. The 10-year period was selected to detecting long-term trends and matched the expert-derived estimate of time over which regular sampling would be required for a monitoring scheme (see Results).
Four large-scale systematic datasets were used to examine the likely range of initial count values and variance parameters, assuming that sites differ by their initial count (random effects intercept) and rate The simulated scenarios differed in (a) initial pollinator abundance (counts per site, from 1 to 200) or levels of pollination service and deficit (from 5% to 30% depending on crop), (b) % decline in pollinator populations over 10 years to be detected (0%, 1%, 5%, 10%, 30% and 50%) and (c) number of sites monitored annually (from 10 to 1,000 sites for pollinators or up to 100 sites per crop for pollination service measures) (see Supporting Information Annex 2, Table A2a for pollinator abundance simulations). These were  Table A2a). The statistical power was estimated as the percentage of simulations that gave a statistically significant result (5% alpha). From these, a detection level of 30% national-scale decline over 10 years (equating to a 3.5% annual decline) was selected as sufficiently sensitive to detect overall population declines, provide sufficient replication to identify plausible drivers of these declines (using additional environmental data such as weather records) and trigger response measures. This was more sensitive than the ideal detection levels recommended across most output measures in the expert survey (see below).

| Costs of monitoring schemes
Costs estimated for each scheme covered (a) staff salaries to undertake field work, identify specimens and administer the scheme; (b) material costs for field equipment, specimen storage, travelling to and from sampling sites and postage of specimens to be identified; However, as inflation affects both the schemes and the estimated costs saved, it is unlikely to affect the conclusions.

| Benefits of monitoring schemes: Research costs saved
Although developed as surveillance monitoring schemes, given the scale of the networks, it is possible to test numerous hypotheses through each scheme by carefully stratifying site selection to capture variance in landscape, climate and management. These potential monetary costs to research funders from using a monitoring scheme as the basis for UK research data collection were estimated by conducting a survey of experts in 2015 to determine the size of the site network and sampling intensity required to address each of eight key pollinator research questions through independent research projects (Table 2).
Research questions were selected from academic literature and policy reports reviewing the key knowledge gaps in pollinator research (Dicks et al., 2013;Vanbergen et al., 2012)  A complete loss of pollination service is unlikely as at least some of the services can be provided by managed pollinators (but see Breeze et al., 2014). The study therefore presents these economic estimates following a 30% loss of pollination, reflecting the cumu- where P it is the price/tonne of crop I in year t, Q it is the total quantity of crop production and ε is the price elasticity of demand; a theoretical metric of the percentage change of price/tonne in relation to a 1% change in total crop production. Φ i is a value equivalent to one minus the dependence ratio (DR) of each crop multiplied by the proportion of pollination service loss (here: 0.3). As there is insufficient data to estimate the price elasticities for all 18 crops, following Gallai et al. (2009), elasticities are set between a low of −0.5 and high of −1.5.

| RE SULTS
Based on the range of initial annual counts per site from the empirical data, the estimated number of sites needed to detect 30% declines with 80% power is either 75 sites where the initial pollinator count per site for a given metric (e.g. bee or hoverfly abundance) is 10 individuals, 145 sites where the initial annual count per site is 1 individual, or 200 sites per crop where initial levels of pollination service and deficit average around 10% (Table 3).

| Costs of monitoring schemes
The estimated 10-year costs of the four potential monitoring schemes varied considerably, ranging from £61,588 for a 75-site volunteer focal floral observation scheme (Scheme 4a) to £26.4 M for an 800-site professional pollination service monitoring scheme (Scheme 2b,

| Benefits of monitoring schemes: Research costs saved
Responses to the expert survey showed variation in median size and regularity of sampling depending on the research question (

| Economic benefits of pollination services
Dependence ratio analysis indicates that pollination services in the  Note: Pollinator dependence = the proportion of yield lost in the absence of pollination. + = taken from a specific UK case study, see Supporting Information Annex 5. Total production = the total market sale price of all UK production of the crop. Pollination benefits = the monetary benefits of crop production theoretically lost with a 30% loss of pollination services. Consumer surplus change = the sum value of consumer welfare changes from rising crop prices, this was performed under assumptions of −0.5 and −1.5 supply elasticity of demand (the % change in prices following a 1% increase in supply). and £435.9 M in annual consumer welfare, 50-164 times the annual costs of the most expensive scheme described (Scheme 2b, 800 sites, Supporting Information Annex 12).

| D ISCUSS I ON
This study is the first to evaluate the full economic costs and benefits of a range of national-level monitoring schemes for insect pollinators and crop pollination services, spanning professional and volunteer-led citizen science approaches. The results demonstrate that a well-designed pollinator monitoring scheme could be a highly costeffective means of addressing key research questions, compared to the costs of implementing separate research projects. A fully professional monitoring network that monitors trends in specieslevel abundance was estimated to save at least £1.96 per £1 spent (Table 6: Scheme 1). The study also illustrates the value of pollinator monitoring as part of efforts to maintain the stability of pollination services to food production. The annual costs of monitoring were estimated at ≤0.006% of the market price of pollinator-dependent crop production and ≤0.02% of the annual economic value of pollination services to consumers lost with a 30% decline in insect pollinators. Although based on the best available data, these estimates are still subject to a number of assumptions that may affect the values estimated but these are unlikely to influence the results overall (see Supporting Information Annex 13 for a review).

| Challenges in implementing nationwide monitoring
Our results are consistent with other work comparing different monitoring scheme structures, with volunteer recorders reducing costs significantly compared with professionals Targetti et al., 2016). However, it is important to consider differences in data quality and taxonomic resolution resulting from different monitoring approaches and how they influence the utility for subsequent management or policy responses (Wintle et al., 2010).
As different species within taxonomic groups provide pollination services to different crops (e.g. Garratt et al., 2016), group-level data species beyond what can be achieved with existing occupancy data (Carvalheiro et al., 2013). Furthermore, although validation of photographic records collected by citizen scientists can improve accuracy , many pollinator species cannot be identified consistently from photographs even by experts (Falk et al., 2019;Morris et al., 2016), although advances in machine learning could theoretically make this more viable in the near future. The only scheme providing a direct measurement of pollination service (Scheme 2: Professional pollination service monitoring) has the potential to detect areas with economically significant deficits . However, the trade-off is that non-crop pollinators, which make up the bulk of pollinator biodiversity (Kleijn et al., 2015) are underrepresented in the data, making it unsuitable for biodiversityoriented objectives (e.g. CBD, 2016;DEFRA, 2015;Protect Pollinators, 2019). In contrast, Scheme 3 (Volunteer collected pan traps) produces species-level diversity data, but if focussed on pan traps and focal observations alone, risks under-representing certain pollinator species due to biases in the species caught (Fijen & Kleijn, 2017;O'Connor et al., 2019). Scheme 1 (Professional pollinator monitoring) would provide the most comprehensive data for baseline surveillance of pollinator biodiversity and (depending on site stratification) enough data on key crop pollinators to potentially act as a valid surrogate for overall pollination services, at the cost of being more expensive to implement. Consequently, although each scheme is capable of answering multiple questions, only Schemes 1 and 3 are likely to deliver a high enough taxonomic resolution for meeting wider biodiversity goals.
Working with inexperienced volunteer recorders also brings additional challenges and opportunities. Volunteers require engagement, including feedback, and training to maintain quality data collection retain them over time (Domrose & Johnson, 2017;Kremen et al., 2011;Mason & Arathi, 2019), they often select sites based on aesthetics or their perceived probability of seeing the focal organisms (Tulloch et al., 2013) and may be unwilling to undertake prescribed methods . Consequently, volunteers may be unsuited to surveying across a stratified, randomly distributed monitoring network, especially in homogeneous agricultural landscapes (Tulloch et al., 2013). We did not explicitly consider the use of expert consultant entomologists as the main data providers because the number of UK consultants with sufficient taxonomic expertise remains below the scale required for the proposed schemes.
This, in turn, highlights the value that a well-designed surveillance monitoring scheme can have in building capacity, particularly among citizen scientists (Birkin & Goulson, 2015;Fox et al., 2017;Gustaffson et al., 2017). As political willingness to act is often driven by public demands (Cardaso, Erwin, Borges, & New, 2011), deepening public involvement and understanding around pollinators, and particularly the role they play in sustaining biodiversity and eco-

| Long-term research benefits
This study is the first to directly evaluate the value of running a monitoring scheme as a form of research infrastructure by comparing the costs of running the scheme to the costs of equivalent primary research. Limited availability of long-term, systematically generated data remain one of the biggest challenges in advancing ecological research (Lindenmayer et al., 2012). Presently, there are only a few long-term ecological projects in the United Kingdom run by research organisations (e.g. Silvertown et al., 2006) or NGO-governmentresearch partnerships (Greenwood, 2003)  were funded, a paradigm of encouraging regular publications, rather than waiting years for more substantial outputs will make it difficult to recruiting qualified staff to lead monitoring work (Lindenmayer et al., 2013).
Key to the success of any monitoring scheme as a research network will be data accessibility, which should be mandatory for publicly funded schemes. Openly accessible long-term ecological data can allow researchers to supplement their own research networks (Fischer et al., 2010), develop cross-disciplinary work (Robertson et al., 2014) and address challenges in theoretical ecology (Lindenmayer et al., 2012).
This study illustrates that a sufficiently large monitoring network can act as cost-effective research infrastructure. In reality, separate power analyses would be required to determine how monitoring networks overlap of with research networks (Buckland & Johnston, 2017); however, the data required for such power analyses would require initial monitoring to establish. Furthermore, the eight research questions considered by the experts are merely an illustrative example, there are many alternative and questions that could be developed within such a future network (Le Feon et al., 2018;Silvertown et al., 2006;Wintle et al., 2010) particularly if integrated into international efforts such as the Long-Term Ecological Research network (LTER Network, 2020). The cost savings presented here are thus likely to be an under-estimate. The methods used in this study could also be expanded to monitor wider insect taxa, particularly when using pan traps which collect non-target invertebrates as by-catch, although care should be taken to avoid an inappropriate 'one-size-fits-all' approach (Lindenmayer & Likens, 2009).
Finally, effective monitoring should aim to underpin a more targeted approach to measuring the impacts of management interventions on pollinator diversity, pollination services and community resilience at a national level (IPBES, 2016;Staley et al., 2016). With better insights into the effects of pressures and management, the question of how much monitoring should be undertaken can be revisited using Value of Information analysis, whereby the costs of monitoring are compared with the likely improvement in management objectives arising from the additional information (Benett et al., 2018).
As there are major deficits in the knowledge base necessary to develop national-scale adaptive management practices for pollinators, monitoring as proposed in this study is valuable in itself (Nygard et al., 2016).
However, as more targeted management becomes possible, there will be resource trade-offs between monitoring and management itself which should be addressed. Nonetheless, such costs are still likely to be far below the value of such monitoring to pollination services and research expertise. Most Value of Information analysis has focused solely on the direct value of monitoring to the management benefits (Bolan et al., 2019). By demonstrating the added value of monitoring as a research tool, our study highlights the potential of monitoring to provide scientific infrastructure that could be funded by research funders while adding value to natural resource management.

| CON CLUS IONS
Policy for supporting pollinators is a matter of international concern (CBD, 2016; IPBES, 2016), yet the systematic data on pollinator abundance and diversity required to determine the impacts of pressures and appropriate responses remains lacking (Garibaldi et al., 2020).
This study demonstrates that even expensive, systematic professional monitoring schemes that deliver the highest quality data can more than pay for themselves. By providing a long-term site network, they can underpin cost-effective research into key questions in a manner not supported by existing research funding, yet which is vital to meeting the UKs international commitments to support pollinator and wider biodiversity conservation (CBD, 2016; Promote Pollinators, 2019).
More fundamentally, by tracking pollinator populations in agricultural landscapes, monitoring potentially allows more targeted and immediate interventions to avoid or reverse economically damaging losses of pollination services (Lindenmayer et al., 2013). This study provides a strong economic and scientific argument that monitoring is both affordable and highly beneficial for ecological research, decision-making, conservation action and ultimately underpinning the transformative change required to sustain nature (Abson et al., 2017).

This work was also supported by the Natural Environment Research
Council award number NE/R016429/1 as part of the UK-SCAPE programme delivering National Capability. The authors thank the 28 experts who took part in the benefits survey, the stakeholders and volunteer recorders involved in discussions around scheme design (in particular, Mark Stevenson of DEFRA) and who provided information on costs of existing schemes or methods.