A real‐world implementation of a nationwide, long‐term monitoring program to assess the impact of agrochemicals and agricultural practices on biodiversity

Abstract Biodiversity has undergone a major decline throughout recent decades, particularly in farmland. Agricultural practices are recognized to be an important pressure on farmland biodiversity, and pesticides are suspected to be one of the main causes of this decline in biodiversity. As part of the national plan for reduction of pesticides use (Ecophyto), the French ministry of agriculture launched the 500 ENI (nonintended effects) monitoring program in 2012 in order to assess the unintended effects of agricultural practices, including pesticide use, on biodiversity represented by several taxonomic groups of interest for farmers. This long‐term program monitors the biodiversity of nontargeted species (earthworms, plants, coleoptera, and birds), together with a wide range of annual data on agricultural practices (crop rotation, soil tillage, weed control, fertilizers, chemical treatments, etc.). Other parameters (e.g., landscape and climatic characteristics) are also integrated as covariates during the analyses. This monitoring program is expected to improve our understanding of the relative contribution of the different drivers of population and community trends. Here, we present the experience of setting up the 500 ENI network for this ambitious and highly complex monitoring program, as well as the type of data it collects. The issue of data quality control and some first results are discussed. With the aim of being useful to readers who would like to set up similar monitoring schemes, we also address some questions that have arisen following the first five years of the implementation phase of the program.


| INTRODUC TI ON
Biodiversity in farmland has undergone a major decline in recent decades (Benton, Bryant, Cole, & Crick, 2002;Donald, Green, & Heath, 2001;Green, 2005;Hallmann et al., 2017;Van Dyck, Van Strien, Maes, & Van Swaay, 2009). Landscape simplification, habitat loss, and the intensification of agricultural practices, including increasing input of fertilizer and pesticides, have been identified as the main causes of this widespread decline (Benton, 2003;Chamberlain, Fuller, Bunce, Duckworth, & Shrubb, 2000;Donald et al., 2001;Stoate et al., 2001). Of the various components of agricultural intensification, pesticide use has been shown to have persistent and consistent negative effects on wild plants, carabids, and birds (Geiger et al., 2010;Hart et al., 2006;Lee, Menalled, & Landis, 2001). While several holistic approaches, such as agri-environmental schemes (AES), have been proposed and implemented to halt and even reverse biodiversity declines (Batáry, Dicks, Kleijn, & Sutherland, 2015;Vickery, Bradbury, Henderson, Eaton, & Grice, 2004), pesticide reduction remains the most important concern, and a primary goal for most European governments and European policies. In this context, and in accordance with the Council Directive 2009/128/EC that established a framework for Community action for the sustainable use of pesticides, the Ecophyto scheme was launched in France in 2008, with the general aim to reduce pesticide use. Several initiatives were put in place, such as training farmers in the responsible use of pesticides, the development of an extensive network of pilot farms to demonstrate good practice ("Dephy farms," Lechenet, Makowski, Py, & Munier-Jolain, 2016), a control program of all the sprayers that are used for the application of phytosanitary products, and the publication of "plant health bulletins" that alert farmers on pest outbreaks so that they can spray only when necessary. However, despite these initiatives, no significant decrease in pesticide use has been detected in France so far (Hossard, Guichard, Pelosi, & Makowski, 2017).
The ways in which agriculture affects biodiversity are multiple.
Agricultural practices influence habitat quality for farmland species, for example, through tillage impacting soil layers and earthworm habitat (Curry, Byrne, & Schmidt, 2002), or fertilization increasing nutrient resource availability and modifying plant community composition in field margins (Fried, Villers, & Porcher, 2018). An example of a well-known but still poorly understood indirect effect is the effect on interspecific relationships (predation or competition) between species and groups of species. Pesticides can thus have direct lethal or sublethal effects on survival or reproduction of plants, invertebrates, and birds (Bohnenblust, Vaudo, Egan, Mortensen, & Tooker, 2016;Kohler & Triebskorn, 2013;Mitra, Chatterjee, & Mandal, 2011), or indirect effects via trophic chains through a decline in plant populations affecting insects and birds (Boatman et al., 2004;Simon, Bouvier, Debras, & Sauphanor, 2010). However, underlying mechanisms remain poorly understood at the population level, despite large numbers of studies. This is particularly true for indirect effects (Benton et al., 2002), which are difficult to study as they require data or experimental studies at numerous trophic levels.
Studying direct and indirect effects of agricultural practices could be approached with multitaxa monitoring, taking into account interactions in trophic chains and estimating population trends. Taxa responses could be estimated by combining taxonomic and functional indices, in order to understand a more complete response to agricultural practices (Chiron, Chargé, Julliard, Jiguet, & Muratet, 2014;Filippi-Codaccioni, Clobert, & Julliard, 2009;Geiger et al., 2010). The potential delayed response of population indicators to pressures and changes, particularly for indirect effects, also requires medium-to long-term monitoring (Aebischer, 1990).
As part of the Ecophyto scheme, a monitoring program was launched in France in 2012, to assess the unintended effects of agricultural practices, particularly pesticides, on farmland biodiversity with a focus on several taxonomic groups that are theoretically not targeted by practices (earthworms, plants, beetles, and birds) at a national scale on 500 different fields. This monitoring program, called the "500 ENI" network (or "500 ENI" monitoring program), has been included in the biological survey of the national territory (Delos, Hervieu, Folcher, Micoud, & Eychenne, 2006 as a legal and regulatory request (art. L251-1 du Code rural et de la pêche maritime). The objective of this monitoring program is twofold: (i) to detect changes in the frequency or abundance of indicator species and simultaneous changes in agricultural practices, including pesticide use, and (ii) to enhance our knowledge in order to create hypotheses on specific mechanisms underlying biodiversity responses (across four taxonomic groups) to agricultural pressures. To achieve these objectives, in addition to monitoring biodiversity indicators, the effects of pesticides must be differentiated from other potential Here, we present the experience of setting up the 500 ENI network for this ambitious and highly complex monitoring program, as well as the type of data it collects.
The issue of data quality control and some first results are discussed. With the aim of being useful to readers who would like to set up similar monitoring schemes, we also address some questions that have arisen following the first five years of the implementation phase of the program.

K E Y W O R D S
agricultural practices, farmland biodiversity, monitoring program, nontarget species factors (e.g., environmental variables such as climate, landscape, sampling conditions such as weather and other agricultural practices such as tillage and fertilization).
Here, we describe the 500 ENI monitoring program and discuss its implementation, an illustration of its potential, and an overview of the dataset. The issue of data quality control is illustrated with the earthworm case, and the first analyses with a botanical example.
Finally, we discuss the challenges that such large scale, long-term, sampling program represents, and the successes and difficulties encountered while implementing it. More in-depth analyses for all taxa and further results will be published in different articles (see , Fried et al., 2018).

| Governance and opening decisions
The 500 ENI monitoring program is managed at national scale by the French Ministry of Agriculture (Direction Générale de l'Alimentation-DGAL). Its structure was based on recommendations by a scientific committee (Comité de Surveillance Biologique du Territoire-CSBT or biological monitoring committee of the territory), and a steering committee (Comité National d'Epidémiosurveillance-CNE or national committee for epidemiological surveillance). The former was mainly composed of researchers from different fields (ecology, ecotoxicology, agronomy) who discussed and decided on the protocols and the variables to be monitored. The latter committee was mainly composed of the French Ministry of Agriculture and agriculture stakeholders, who decided how to implement the monitoring program in France.
The 500 fields are distributed across the whole metropolitan France (including Corsica) in order to represent the different landscapes and pedoclimatic contexts of the country. The monitoring is focused on four crop types representative of the main crop systems in France and contrasting production system (e.g., organic versus. conventional) to take into account contrasting impacts on biodiversity (Hole et al., 2005;Kragten et al., 2010). The four crop types include at some point of their rotation, either: (1) annual crops, including (a) winter wheat (189 fields) and (b) maize (155 fields); (2) vineyards (99 fields), or (3) market gardening crops (57 fields). Crop types and number between parentheses refer to the number of fields grown with this crop at the launching of the monitoring in 2012 ( Figure 1). As the fields are fixed, the crops change each year except for vineyards and some fields with monocultures (e.g., maize).
Market gardening, although it is associated with several crop species per field within a year, remains a coherent group with crop species differing from those grown in annual field crops. Hence, fields with lettuce are considered as a market gardening crop model, as lettuce is one of the most widely cultivated vegetables in France, across all regions and seasons. In annual field crops, each field is characterized by a rotation of different crops observed through several years of surveys, with, however, a dominance of winter crop species in fields categorized as "winter wheat" and a dominance of spring crop F I G U R E 1 Distribution of monitored fields at least one year between 2012 and 2018 in metropolitan France. Purple: vineyards (n = 104), green: lettuce (n = 55), brown: winter wheat (n = 219), yellow: maize (n = 151). For the annual crops (lettuce, winter wheat, and maize) refers to the crop planted at the launch in 2012 species in fields categorized as "maize." Although the "reference crop types" are named with a single crop within a rotation, the four reference crop types still represent distinct crop rotations. Fields with vineyards are considered as the perennial crop model. Each French region (an administrative unit that divides France into 22 areas averaging 25, 000 km 2 ) decided on the local implementation of the program, under national recommendations. The sampled fields and within-field observation sites are chosen such that long-term trends in biodiversity indicators may be derived within the fields and field margins. We believe that unintended effects should be measured in the field margins and not only inside the fields. To assess the unintended effects of agricultural practices, the taxonomic groups monitored are mainly nontarget species (in respect to pesticides) that are sensitive to agricultural practices (Cluzeau et al., 2012;Pérès et al., 2011). Four taxa (earthworms, wild flora, beetles, and birds) were selected to represent the different compartments of biodiversity in the agro-ecosystems and to cover different spatial and temporal scales of responses to landscape and agricultural practices. Agricultural professionals, such as farmers' advisors, are in charge of the data collection under supervision by the regional delegations of the Ministry.

| Study sites and sampling design
Approximately 500 fields were surveyed every year since 2012, representing a total of 523 unique fields between 2013 and 2016. A stratified sampling was chosen. Each French region was allocated a number of fields consistent with the importance of the focus crops at the regional scale. We selected 80% of fields under conventional farming and 20% under organic farming in each region, to ensure a good representation of organic farming, which represents only 8% of the utilized agricultural area in France. The fields also had to be located within existing small agricultural areas (a zoning dividing France into 713 homogeneous agricultural areas, considered to be relevant for agronomic issues) and avoid infrequent or unusual agricultural practices.
To ensure the representativeness, we verified for each focus crop that the proportion of fields selected in each region is correlated with the proportion of the focus crop in all the regions, based on national agricultural statistics (see Appendix S1). The average proportion of noncropped area around the fields surveyed is also correlated in each region with the proportion of natural elements in the landscape, using the High Nature Value index HVN3 (Pointereau, Paracchini, & Terres, 2007).

| Biodiversity surveys
Flora and coleoptera are surveyed in field margins, one of the most important refuges in agricultural landscapes for wild flora and fauna (Marshall & Moonen, 2002). Field margins may be considered as seminatural landscape elements, which are nevertheless highly exposed to the unintended effects of agricultural practices. Earthworms are sampled within the fields because of their low capacity for dispersal, while birds are recorded at a larger spatial scale including both the field and the adjacent area.
Observers were selected by the regional organizations. Except in rare instances, most observers are nonexperts in respect to the four taxa. Existing standardized protocols, with demonstrated reliability, have therefore been simplified to permit their application by nonexperts. To this end, experts created a shortlist of the most common bird and plant species, including both farmland specialist and generalist species, that have a broad geographical distribution and offering a wide range of ecological requirements to enable trait-based comparisons. For beetles and earthworms, all sampled specimens were collected and classified into morphological groups, assisted by determination keys.
Earthworms are considered to be ecosystem engineers (Jones, Lawton, & Shachak, 1994) as they modify the physical, chemical, and biological parameters of the soil, thereby providing ecosystem services (Blouin et al., 2013). Moreover, they are considered one of the most accurate biological indicators of soil quality due to their sensitivity to soil characteristics (Lee, 1985), land cover (Ponge et al., 2013), and pesticides (Pelosi, Barot, Capowiez, Hedde, & Vandenbulcke, 2014), all of which may impact earthworm abundance, functional structures, and species composition (Paoletti, 1999;Pérès et al., 2011). They also represent an important food source for many farmland specialist organisms such as birds or insects (Edwards & Bohlen, 1996;Lavelle et al., 2006). Earthworms are sampled with the mustard method (Gunn, 1992) adapted by the French Participatory Observatory of earthworms (OPVT, https://ecobi osoil.univ-renne s1.fr/OPVT_accue il.php). Earthworms are sampled once a year in spring, in three 1-m 2 replicate quadrats located 6m apart inside the field. Each replicate is watered twice with ten liters of mustard solution. Earthworms that emerge on the surface in response to the irritant solution are collected, counted, and sorted into developmental stages (adult or immature) and functional groups: epigeic (EPI), epi-anecic (EpA), anecic strict (AnS), and endogeic (END), see Bouché (Bouché, 1972(Bouché, , 1977. To ensure access to the raw data and to allow identification to be verified if necessary, all samples are photographed and kept in alcohol. They are subsequently sent to the ECOBIO laboratory to be stored as a data reference and for identification to subspecies level (see Box 1).

| BOX Earthworm survey
Mustard protocol. The Mustard Protocol is a standardized protocol.
The goal of this French Participatory Observatory of earthworms (OPVT) protocol is to compare earthworm abundance in respect to different types of management and soil cover, etc. In order for comparisons to be possible, sampling conditions have to be very similar.
• Sampling must occur in spring during the period of earthworms' activity (from January to April), preferably in the morning and before any tillage, or 4 weeks after tillage.
• The soil must be wet, but not sodden. Soil temperature must be between 6 and 10°C.
• Sampling must be in a flat, homogeneous area which is representative of the whole field. In order to avoid a border effect, the positioning of the sample areas must be more than 10 m from the field margin.
• All the observers must use a standardized sampling kit.
Three areas of 1 m 2 are delimited, spaced out by 6m on a homogeneous surface that is representative of the field. The vegetation in the square meter and 10 cm all around is cut as short as possible and removed from the sampling area ( Figure 2).
2. Mustard solution is prepared on site. The ground is watered twice in each delimited area. For each treatment (6 in total per sampling session; 2 per delimited area), 300 g of Mustard "© Amora fine et forte" is thoroughly diluted in 10 L of water.
3. The mustard solution is spread across each 1-m 2 sample area.
After 15 min, the second treatment is applied. All earthworms Monitoring flora only in field margins was justified due to the fact that field margins contain the greatest botanical diversity in intensively managed lowland landscapes and represent a refuge for many plant species (Marshall & Moonen, 2002). Field margin flora are also an important lower level of the trophic chain, providing a resource exploited by both insects and birds (Marshall et al., 2003) as well as providing a habitat for a range of fauna, including small mammals. Finally, as opposed to weeds within the field, plant species in field margins are not directly targeted by herbicide treatments and other weed control practices, so that the impacts of farming practices on these plant communities are genuinely unintended. Plant species are identified in ten 1-m 2 quadrats located in the field margin strip, sensu Marshall and Moonen (2002), which runs between the surveyed cultivated field and a distinct neighboring area (which may be a road, ditch, track, hedge, another cultivated field, or another habitat). Once per year, at the peak of the flowering period (June in most cases, but may be from April to August depending on the latitude of the site), the presence/absence of plant species is identified in the ten quadrats, to produce a frequency of occurrence (1-10) for each species present for each field margin. Observers are expected to be able to identify a list of 100 focus plant species; however, all taxa found in the quadrats must be categorized; that is, species not belonging to the focus list must be identified to genus or family, or to any superior taxonomic level, provided that all the different taxa present in the quadrats are distinguished. Each year, an expert verifies the data for species distribution and phenology. More detailed information on the protocol, including the area surveyed, the layout of the ten quadrats, the sampling period, and the selection criteria for the focus species, is given in Box 2 (Greaves & Marshall, 1987;de Snoo & van der Poll, 1999).

| BOX Field margin vegetation survey
Observation area. The aim of the flora survey is to detect unintended effects of farming practices on nontarget organisms within field margins. Three primary areas are recognized in field margins (Greaves & Marshall, 1987): the crop edge, the field margin strip, and the field boundary. The area surveyed in this study excludes the crop edge (also called the conservation headland), which is located within the 1-6 first meters of the crop ( Figure 3). It also excludes the cultivated strip, immediately outside the last row of crop, which is an area with primarily bare soil, and is usually colonized by weed species from the field. The area of interest in this study is the field margin strip, which is the area of herbaceous vegetation between the cultivated strip and the adjacent landscape (see Figures 3 and 4), the latter being either another cultivated field, a road or a track, another habitat (grassland, forest), or a field boundary (hedge, fence).
Species surveyed. Observers are expected to be able to identify a list of 100 focus plant species; however, all species found in the quadrats must be individually identified to an appropriate taxonomic level, that is, species which do not belong to the focus list may be identified at the genus or the family level, or at any superior taxonomic level, provided that all the different taxa present in the quadrats are distinguished.
In 2012, four draft lists of 50 focus plant species were established according to the crop-type and the regions (Mediterranean regions versus non-Mediterranean regions). These lists comprised species that are more representative of agricultural landscapes, known as "agrotolerant" species, and species that are more representative of the perennial herbaceous vegetation of the field margin strip, or of adjacent natural or semi-natural habitats ("nature-value" species). These two groups were included as their responses to disturbances related to farming practices are expected to be different.
Based on previous studies on field margins (de Snoo & van der Poll, 1999), the selection of the focus species was also determined by species traits, in order to have a broad representation of both broadleaved and grass species, annual and perennial species, and of plant species pollinated by insects, self-pollination, or physical agents such as wind; and finally of species that are highly responsive to nutrient availability (e.g., nitrophilous species), and species that are tolerant of poor soil conditions (oligotrophic soils).
Based on a preliminary survey conducted on the 500 field margins in 2012, the draft lists of 50 species were supplemented with all the species that were recorded in more than 5% of field margins, resulting in an addition of about 30 species. In order to round off the focus list at 100 species, additional species were added, which were present in 1%-5% of the field margins, and corresponded to the above criteria. All of the species on the focus lists are relatively common species (or were common up until recently) and may be found across the entirety of metropolitan France. However, the focus lists are slightly different between annual crops (wheat, maize, lettuce) and vineyards because the latter are often found on exposed hillsides, with a different range of species than those found in planes or valleys. Finally, the focus lists also differ distinctly between the Beetles are also a food source for birds (Clere & Bretagnolle, 2001;Green, Tyler, & Bowden, 2000) and otherwise herbivorous species, which makes them interesting for the study of trophic links. Experts also feel that they have a closer link to crop types than other insect groups such as butterflies or bees, because of their typically more

| BOX Beetle (Coleoptera) sampling protocol
The aim of this protocol is to survey the coleoptera present in the herbaceous field margins. This protocol is adapted from an entomological study conducted over 6 years in the Ile-de-France region by Chauvelier and Manil (2014).
Protocol overview. Sampling is carried out within the field margin strip, which is the herbaceous area of vegetation between the cultivated strip and adjacent patch in the landscape (see Box 2), the latter being either another cultivated field, a road or track, another habitat (grassland, forest), or a field boundary (hedge, fence). The transects are placed in the centre of the field margin strip, as in the flora protocol.
Beetles are sampled by sweep-netting 3 times in spring between April and July, in order to include several periods of emergence.
During each visit, observers collect all the beetles on 2 transects of F I G U R E 2 Sampling instructions for the earthworm survey 20 steps and 40 sweeps (20 double sweeps: at every step, the operator moves the net from left to right and then back, perpendicular to the walking direction). Adult beetles are collected with a vacuum cleaner from the sweep-net before they escape and are then killed and stored in alcohol for preservation. Beetles are then promptly classified into 14 morphological groups (Chauvelier & Manil, 2014b).
The protocol includes a key to assist observers in classifying the beetles, and a guide with a general description of each morphological group. The abundance of each morphological group is then recorded. Beetles are photographed on graph paper, which provides an internal scale, in order to store the raw data and use it for identification control.
Classification. This protocol and the determination key were developed by Claude Chauvelier, an entomologist specializing in beetles, based on his field experience. The fourteen morphological groups (Table 1) correspond approximately to the principal families of beetles. The determination key is very simple compared to conventional methods for identifying beetles and allows nonexperts to classify all specimens. A miscellaneous group is also included for those specimens that have not been classified into any of the groups described by the key.

| BOX Bird survey
The aim of the protocol is to survey the common bird species around the fields. This protocol is a simplified version of the French Breeding Birds Survey (FBBS) (Julliard & Jiguet, 2002), allowing the application by nonspecialists. Species surveyed. Observers are expected to be able to identify a list of 28 focus bird species (see Appendix S3), by sight or by ear.
They must also be able to distinguish other species, in order to differentiate them from the focus species.
Species on the list have been selected in order to account for a range of ecological specializations, diet, habitat preferences, and distribution, applicable to the entirety of metropolitan France and to different crops. The most common species in farmland habitats were chosen through the analysis of the FBBS database and regional reports, and consultation with experts. The selected species have broad distributions (common both in farmland and across the majority of metropolitan France), relatively close links to the target crops (i.e., are known to breed in arable fields or vineyard), and include both specialist and generalist species (Jiguet et al., 2007 The number of 28 species has been judged by nonskilled observers as a maximum they can learn.
We expect this list brings a good representativeness of species and that their responses to disturbances related to farming practices in the field would be different as their trait and links to crops are different.
A specific focus list has been proposed for Corsica as the communities are rather different from those of the continent.For each of the protocols, a notable effort has been made to standardize the observations through training, materials, format, etc.
An example of an important challenge to this standardization is the consistency of equipment for the earthworm and beetle protocols.

| Landscape variables
Landscape composition and configuration are major determinants of biodiversity (Burel, Butet, Delettre, & Millàn de la Peña, 2004;Burel et al., 2013;Fahrig, 2003). There are numerous ways to assess landscape variables, and their application can be complex (Li & Wu, 2004). The choice was made to reduce the number of indices by focusing on simple, complementary, and understandable metrics. A preliminary study was conducted to describe the landscape around the 500 monitored fields. Each field was georeferenced in a GIS database using ArcView (ESRI, 2000), and the sampling zones were mapped on aerial photographs. We used two landscape databases of land cover and land use in France: BD Topo (topographic database) and BD Parcellaire (administrative field database) from the IGN (National Institute for Geographical and forest information).
In addition, we used the "Registre Parcellaire Graphique" database 125, 250, and 500 m. This process would need to be replicated for new fields integrated to this program to compensate for those that could be lost.

| Agricultural practices and additional variables
Through their shaping of the landscape, agricultural practices are the main determinant of biodiversity trends and composition in agricultural habitats (McLaughlin & Mineau, 1995). In order to take into account all aspects of these practices, including pesticides, which may influence the 4 biodiversity indicators, a wide range of detailed information was collected through yearly interviews with the farmers who owned the monitored fields. Each interview lasted between 1.5 and 4 hr (sometimes more), depending on the type of interview (on the phone or at the farm), the amount of interventions carried out on the fields, and whether or not the farmer had prepared the interview (by gathering the necessary information in advance). Field data collected include several categories of detailed variables.

| Data management and access
All data, including the yearly monitoring of the four taxa as well as all other relevant information on agricultural fields, are uploaded by observers with a purpose-built application and stored in a PostgreSQL database (version 9.1). At the end of 2012, which was the first year of the 500 ENI network, adjustments in the protocols were made, and details of agricultural practices were added to the collected variables. In addition, observers were trained on the protocols F I G U R E 3 Different types of field margin strips along a farm track (on the left) or next to a field boundary delimitated by a hedge (on the right) throughout 2012, so the data collected during this year were affected by the initial noncompliance with protocol conditions. 2012 is therefore excluded from our analysis.
Currently, the database is available by request for French public research. All requests must be addressed to the General directorate for the food and plant health section (DGAL), Ministry of Agriculture of France, Paris.

| Statistical methods
Because the selected fields are spread over the entire metropolitan territory, an important portion of the variation in biodiversity indices is expected to be explained by nonagricultural factors, such as landscape composition and structure, pedoclimatic contexts, and the daily conditions during observation. It is also expected that selected taxa display differentiated responses. The first stage of analysis will be to identify influential variables that impact the biodiversity structure of each taxa.
To assess the impact of agricultural practices on biodiversity, it is necessary to disentangle different sources of variability at regional or local scales, as well as to consider different ways to build diversity indexes from raw observational data. Dependent variables may simply be common species counts (richness index) for birds or flora, counts at higher taxonomic level as for beetle or earthworm data, community composition, or functional diversity. Statistical methods that will be applied to the dataset range from linear model (LM, typically multiple regression), generalized linear model (GLM), when the variable type (counts) generates non-Gaussian residuals (Zuur, Ieno, Walker, Saveliev, & Smith, 2009), to generalized additive model (GAM) when the links are potentially nonlinear and in a mixed framework, for example, to take into account different sources of pseudo-replication when relevant (Pinheiro & Bates, 2000) with LMM, GLMM, or GAMM (Zuur, Saveliev, & Ieno, 2014). We also intent to consider recently developed statistical methods such as Bayesian ordination, to perform multivariate multiple regressions including latent variables that would allow to deciphering the contributions of environmental conditions and biological interactions in shaping community structures (Hui, 2016;Warton et al., 2015). Raw data from repeated surveys at the same sites are temporally autocorrelated due to the site effect. Surveys therefore cannot be considered as independent surveys, and statistical modeling must account for this, either by evaluating residual's independence after modeling site fixed effects and temporal trends, or by considering a random effect. If spatial and temporal correlations of the residuals constitute a problem, we must use additional covariates or a term defining the structure of correlation for the residuals. Explanatory variables include several forms of data from quantitative variables such as the total amount of nitrogen (N) in fertilizers, categorical variables such as type of fertilizer, or binary variables such as organic or not. Predictor variables may also be grouped into three types: variables that characterize the observational conditions (e.g., date, time, weather, and observers), variables that describe the physical environment (e.g., landscape characteristics), and finally, variables characterizing agricultural practices in the monitored fields or in the field border. With more than one hundred potential predictor variables, specific care must be taken to be careful of any missing data. If all previously applied statistical methods require full records, it becomes necessary to use missing data imputation or alternative methods to avoid eliminating too much of the annual data.
To analyze community composition, which is in some cases more informative than a standard biodiversity index, we propose the use of multivariate methods such as (partial) canonical correspondence analysis (CCA) (Ter Braak, 1986) or (partial) redundancy analysis (RDA) (Van Den Wollenberg, 1977).

| RE SULTS
The result of the initial implementation of the 500 ENI network is presented in three parts in order to outline its potential, beginning with a description of the dataset, the issues of quality control for the earthworm data, and finally some preliminary results for the botanical data.

F I G U R E 4
Layout of the 10 quadrats in the field margin strip

| Overview of the biodiversity dataset
Between 2013 and 2016 (4 years), 12,888 biodiversity surveys (defined by field, date, and protocol) were conducted on 523 fields, carried out by 338 observers across metropolitan France and Corsica (Table 3).
For earthworms, among the four functional groups, endogeic have the highest abundance per m 2 (7.8 ind/m 2 ) and epigeic have the lowest abundance per m 2 (1.9 ind/m 2 ). Except for epigeic, the proportion of juveniles is higher than for adults, particularly for epi-anecic (72% juveniles) ( Figure 6). For beetles, the three most frequent groups captured were leaf beetles (Chrysomelidae), weevils (Curculionoidea), and ladybugs (Coccinellidae), all three groups being observed in more than 75% of fields and 50% of surveys (Figure 7). The first two groups are mainly phytophagous, and the third one is primarily predatory. The miscellaneous group is composed of several families of beetles. Relative frequencies between years (2013-2016) seem to be generally stable; however, the absolute abundance is very variable.
The most frequently observed bird species from the survey list were the skylark (Alauda arvensis), the wood pigeon (Columba palumbus), carrion crow (Corvus corone), and blackbird (Turdus merula), all observed in more than 50% of fields every year and in more than 30% of all surveys (Figure 8). Among these, the skylark is a farmland specialist species widely distributed in France; the others are generalist species, also with a widespread distribution. This first descriptive exploration of the data also highlights that some species such as the harrier, despite being widely distributed, easy to identify, and strongly associated with agricultural field (i.e., farmland specialists, e.g., breeding in crop fields), could not be used in our analyses because of their scarcity in the dataset (<5% in frequency).

| Overview of the dataset on agricultural practices
The

| Quality control
For beetles and earthworms, identification was verified by experts using photographs (beetles) or from preserved specimens (earthworms). This verification revealed gaps in the observer training that had led to errors. Remedial action has been necessary, such as the development of tools to improve classifications (an identification key, a training quiz, etc.) and further training sessions.
An example with earthworms is used here to illustrate data qual-

| Initial analysis on the impact of agricultural practices
In order to illustrate the first analyses conducted, some of the early results for flora are presented, based on 430 field margins surveyed in 2013 and 2014.
With an average of 16.5 ± 6.4 species, organic field margins were significantly richer than conventional field margins, which had an average of 14.1 ± 6.6 species (Student's t test, t = 3.690, p < .001).
Preliminary analyses show a relationship between field margin plant community and pedoclimatic gradients, field margin management, and in-field practices (fertilization), while species richness depends more upon field size and management intensity (intensity of herbicide use) (see (Fried et al., 2018) for further details).
Within the farming practices that showed a significant impact, there was notably a slight positive correlation between the amount of   within the field that effectively selects for species within field margins that are known to be the more nitrophilous. Thus, nitrophilous species such as Urtica dioica, Poa trivialis, Elytrigia repens, or Plantago major were more frequent in field margins with high-nitrogen fertilizer input, while Trifolium repens, Vicia sativa, Achillea millefolium, and

CARABIQUES-ground beetles
Erodium cicutarium were more frequent in field margins with low or no nitrogen fertilizers.

| D ISCUSS I ON
The massive use of pesticides by intensive agriculture during the last few decades has become a major threat to the persistence of biodiversity in agricultural landscapes. Several studies have investigated the role of pesticides in the decline of farmland bird species (Eng, Stutchbury, & Morrissey, 2017;Mineau & Whiteside, 2013). In the context of these results, a large-scale monitoring program has been launched in France on both species abundance (or richness) and agricultural practices, to better assess the direct and indirect effects of farming practices on biodiversity, as characterized via four taxonomic groups (earthworms, plants, beetles, and birds). Generally, such programs focus on regions of interest (Bretagnolle et al., 2018), for example, the Natura 2000 Special Area of Conservation (SAC) (Brodier, Augiron, Cornulier, & Bretagnolle, 2014), or on target species or small communities. Nationwide citizen science programs also produce beneficial and widely acknowledged data on indicators of the state and responses of biodiversity (Chandler et al., 2017), and have contributed to demonstrating the decline of biodiversity in agriculture (Jiguet et al., 2012;Mineau & Whiteside, 2013). However, citizen science studies were unable to specify the potential causes of the decline, because of the lack of accurate data on the types of agricultural practices around observation sites. In addition, volunteer-based programs encounter bias due, for example, to a large turnover of both observers and observations sites (Boakes et al., 2016), and also due to many sites being monitored only once (Hallmann et al., 2017). At the same time, targeted studies on pesticide impacts

TA B L E 2 (Continued)
and multispecies interactions are based on experimental, controlled, or semicontrolled conditions (Tremblay, Mineau, & Stewart, 2001) at a relatively limited spatial and temporal scale, and often focus on a small number of pesticides or practices to reduce variability (Chiron et al., 2014). Likewise, risk assessment studies on preapproved pesticides are unrealistic (Dalkvist, Topping, & Forbes, 2009) because toxicology is more focused on individual responses than on ecological or population impacts (Pelosi et al., 2014;Schmolke, Thorbek, Chapman, & Grimm, 2010;Sibly, Akçakaya, Topping, & O'Connor, 2005). Nationwide monitoring programs that include a large selection of species from different taxonomic classes and focus on agro-ecosystems are relatively rare. The French 500 ENI program is at the moment unique in Europe in the respect that it consists of long-term surveys across France at fixed sites, collecting detailed data on agricultural practices under real-use conditions as well as on the abundance of species or species groups in four taxonomic classes. In addition, for flora, birds, and insects it focuses on the field margins, while most agro-ecological studies target field interiors (Fried, Kazakou, & Gaba, 2012). Nationwide and global monitoring programs are essential for assessing how public policies ensure the conservation of biodiversity and associated ecosystems services in agricultural areas, especially when the reduction in both pesticide use and of biodiversity loss is a major target. This long-term study is complementary to the ones that focus on the specific mechanisms that lead to indirect effects (Hart et al., 2006), but the latter studies are not a substitute for a large-scale investigation.

| PRELIMINARY RE SULTS: LE SSONS FORM THE PROJEC T
While long-term monitoring programs are demonstrably important, their implementation is not necessarily easy to achieve. Many difficulties appeared at the beginning of the program, or in the early stages that follow. For the current study, some difficulties have continued to persist five years after the launch, and these challenges must be taken into account throughout the duration of the program.
Based on our experience, we propose to classify the challenges inherent to this type of program into three broad categories and in order of difficulty, for the purpose to aid in the future avoidance of such challenges.

| Basic advice and easily avoided challenges
• Establish a committee of experts to choose or develop protocols and select relevant agricultural practices and taxa to survey.
The role of the committee is to ensure that the consideration of practices, landscape, taxa, and socioeconomic aspects is comprehensive. To this end, the members of the committee require skills in assessing the feasibility and implementation of complex programs. Time will also be required to resolve conflicting opinions or objectives and to establish a consensus.
• Decide on the sample size for the numbers of fields and farmers monitored. This is intended to achieve the optimal trade-off between total cost and the ability of the survey to detect trends, changes, and regional effects. Before the program begins it is useful to carry out sensitivity tests to determine the number of sites necessary to obtain sufficient resolution of the ecological trends.
• Consider the geographical scales investigated during the study, from local impacts to regional effects, including intra-field  in biases in the analyses. However, higher taxonomic levels that involve too much aggregation will also impair future analyses by limiting their capacity to detect effects beyond the most general trends.
• Anticipate the statistical treatments/analyses to be performed.
This requires the relevance of the objectives and the completeness of the collected data to be verified. Potential biases that may be created when strict compliance with the observation protocols is not possible must be modeled and corrected from the records of observed conditions (e.g., the time after sunrise, or presence of rain during bird surveys). Similarly, stratified sampling schemes should also be taken into account in order to predict diversity changes at a national scale. In order to improve the analyses, external data that is accessible at low cost may provide useful complementary observational data (e.g., remote sensing, pesticide sales by region, and crop areas by region).
• Anticipate that the data produced during the first year will probably be very incomplete, and that during the first two or three years the protocol will be adapted several times in response to the characteristics of the initial results. The data from the first few years will therefore be difficult to integrate into subsequent long-term studies.

| Difficulties and compromises discussed by program participants
• In order to avoid inaccuracies and misunderstandings in the field data, the questionnaire on agricultural practices must be • Consider who the observers should be. This role involves data collection on agricultural practices, taxonomy, and environment.
This requires a wide range of skills, so compromise is necessary.
Training is also necessary, and a good knowledge of the local context is preferable. These choices have consequences on the quality of the data and determine which environmental effects will be identifiable. is also greater conservation value in organic field margins (Aavik & Liira, 2009;Bassa, Boutin, Chamorro, & Sans, 2011). It is likely that this is due to reduced herbicide drift and mineral fertilizer runoff in organic fields. A study based on the early data collected in 2013 and 2014 has been published for field margin flora, using a functional approach linking species traits directly to environmental variables including agricultural practices (Fried et al., 2018).

| Difficulties present in all sections of the program
However, for most taxa, four years of survey seem to be too limited to identify all types of ecological responses to agricultural practices, the interactions with varying environmental conditions, and to detect significant changes in long-term trends in species abundance or communities' composition. This kind of program must be viewed through a long-term perspective and will only show all its full potential with time passing, especially in respect to temporal trends where a decade of data seems to be necessary in order to account for the high temporal variability of some taxa, or to include species that are still too rare in the present dataset.

| REG UL ATORY PER S PEC TIVE S
The detection of an effect of pesticides or another practice on one of the 4 biodiversity indicators (earthworms, plants, beetles, agriculture, food, and forests). This scheme aims to detect and prevent the risks associated with pesticides, potentially resulting in either the withdrawal of an approved pesticide or the imposition of constraints on its use until causality is confirmed.

| CON CLUS ION
The French 500 ENI program is an ambitious project dedicated to monitoring the effects of agricultural practices on biodiversity ( the French Agency for Biodiversity (Agence Française pour la Biodiversité). We would like to thank everyone that has collected data in the field, the farmers who provided information on their practices, and everyone involved in the coordination of the data network. We thank Hoël Hotte, for his participation in the coordination of the network for the earthworm indicator. We would also like to thank Marie Carles and Sylvie Ladet (INRA) for providing the landscape composition metrics.

CO N FLI C T O F I NTE R E S T
There are no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
All the database is available by request for public research. All requests must be addressed to the General directorate for the food and plant health section (DGAL, bsv.sdqspv.dgal@agriculture.gouv. fr), Ministry of Agriculture, Paris.