We analyzed the disparity between the actual distribution of sites from this bottom-up in situ program and a stratified sampling across environmental and geographical gradients. Such balanced sampling is anticipated to catch a broad spectrum of the intra-specific genetic diversity (Tobón-Niedfeldt et al., 2022). For this objective, it was essential to determine the potential distribution of the candidate meadows throughout Switzerland’s biogeographical regions and agricultural zones. To map out the complete potential distribution of the candidate sites within the program, we used the shapefiles of the agricultural area in Switzerland (FOAG, 2023). We retained only areas corresponding to permanent meadows, pastures and wooded pastures i.e. areas with codes 613, 616 and 625. These categories are the only eligible for the in situ program. Moreover, we excluded ECAs from the layers as they belong to other subsidy instruments ( Supplementary Figure S1 ).

The GIS layers were sourced from the Federal Office of Agriculture in April 2023. For faster data manipulation, vector layers were rasterized to a 10 m resolution using the terra library (Hijmans et al., 2023) in the R software, version 4.0.3 (R Core Team, 2020).

To delineate different strata along biogeographical and altitudinal gradients, we overlaid the map of Switzerland’s biogeographical regions (FOEN, 2022) with the map of its agricultural zones (FOAG, 2023). Biogeographical regions are delineated based on shared similar ecological characteristics and history. However, they lack a finer subdivision along the altitudinal gradient (FOEN, 2022). For this, the distinct agricultural zones of Switzerland are useful as they are subdivided based on altitude levels, climate, transport routes and topography. They encompass seven zones, ranging from the lowlands to the alpine levels (FOAG, 2023). Out of 84 theoretical units (12 bioregions * 7 agricultural zones), we identified 66 biogeographical strata, as some of the strata were not present across all the biogeographical regions ( Supplementary Figure S2 ).

Therefore, we could identify the potential area for the in situ program in each stratum ( Supplementary Table 1 ). Balanced stratified sampling assumes an equal sampling effort within each stratum. However, some strata have only very few potential areas ( Supplementary Table 1 ). For strata with a potential area equal to or lower than 420 ha, we assumed that only one-tenth of these small areas could be sampled. Therefore, a proportional sampling was assessed in these small strata.

For each association, the list of characteristic, dominant or companion species was sourced from Dietl and Jorquera, 2015; Eggenberg et al., 2015 ( Supplementary Table 2 ). For each of these species, a potential distribution map was derived from species distribution models (SDMs) linking species observations to topo-climatic layers (Guisan and Thuiller, 2005; Elith and Leathwick, 2009). We used the same methods, data and environmental datasets as described in Petitpierre et al., 2023. To summarize, species with enough occurrences (n ≥ 10) were associated with an initial set of 33 environmental layers describing topography, climate and remote sensing factors. An initial selection process retained only the most relevant variable, keeping between two and nine variables, depending on the species (see Appendix 1 in Petitpierre et al. (2023)). These variables were then related to species occurrences by combining three modelling algorithms (general additive models, MaxEnt and gradient boosting model) into an ensemble modelling approach (Thuiller et al., 2004; Araujo and New, 2007), or an ensemble of small models (ESM, Breiner et al., 2015) depending on the number of observations. This approach produced continuous suitability maps for each species at a resolution of 100 m. These continuous maps were reclassified into binary maps (i.e. potential presences and absences) by setting a suitability threshold to obtain an omission ratio of 10%. The map of the potential distribution of each association was then obtained by stacking the potential distribution maps of the species contained in the association. In the stacking, maps of characteristic species were given a weight of 1.2 while the other species had an assigned weight of 0.2, i.e. five times less than the characteristic species. Therefore, the stacking of the species potential maps of each association produced a map of a continuous association score. This score was also reclassified into a binary score (i.e. favourable or unfavourable) by comparing the potential distribution of the association with the observed distribution of the associations in the InfoFlora database (Infoflora, 2023), which is the competence centre for information on the wild plants of Switzerland. It contains more than 11’000’000 observations of the Swiss flora and associations score were calculated at a 100 m resolution for Switzerland. The continuous association score of the potential maps of the associations was binarized by setting a threshold to obtain an omission ratio of 5% regarding the observed distribution of the associations, except for Poion alpinae where it was set to 7.5% for a more plausible distribution, according expert’s opinion.

To obtain a more exhaustive list of plant species located at each site of the in situ program ( Supplementary Table 3 ), we extracted the observations from the InfoFlora database in a radius of 75 m around the coordinates of each site and kept the observations of the priority CWR of Switzerland. Only occurrences observed between 01.01.2002 and 31.08.2022, and with a coordinate uncertainty smaller than 150 m were retained. Importantly, as the observations of the InfoFlora database consist of different types of datasets (opportunistic observations, monitoring, surveys), more species can be present around the sites and not being reported in the database.

Taking the actual stand of deployment (57% quotes filled), the next point is, therefore, to identify where the next surfaces should be ideally situated to significantly improve the conservation of target species and associations. To what extent would the 1’566 hectares included in the program provide sufficient coverage? And if not, how to incentivize the other cantons or farmers to participate? Is the auction-based system the more appropriate scheme when the offer is not sufficient to fill the quotas? Two years after inception, 1’217 surfaces could be integrated into the in situ program, covering a total of 1’566 hectares ( Figure 1C ). This covers 57% of the allocated quota (2’750 ha) distributed in 23 out of 26 cantons. Some cantons simply did not find any suitable surfaces in the first round, other were reluctant or even refused to participate, as the program was judged not worthy in terms of costs/benefits balance. While the modalities of the in situ program are by definition bottom-up, there is no direct control over the localization of the surfaces being integrated, nor their floristic quality. The relative quantity of each association within the allocated surfaces is only monitored indirectly.

Considering the surfaces already integrated into the program ( Figure 1C ), the sites are distributed across 49 biogeographical strata (out of 66; 75%). The potential area for the eligible sites is bigger than 100 ha in 9 empty strata ( Supplementary Table 1 ). We can observe a bias in surfaces considered in the in situ program towards the East of the country ( Figure 1 ). A balanced stratified sampling of 2’750 ha would tend towards 51.63 ha in each stratum with a potential area ≥ 420 ha and 10% of the potential area in strata with a smaller potential (surfaces for each biogeographical unit ( Figures 1A, B ; Supplementary Table 1 ). The current distribution of the sites is only weakly correlated to such a balanced stratified sampling (Pearson’s correlation = 0.33, p-value = 0.007) with a significantly different distribution (Chi-square test p-values <0.001). To reach a distribution close to a stratified sampling, the next sites should in priority target the Prealps, Central Western Alps and Southern Ticino regions with about 150 to 190 ha to add, while Rhine Basin, East Central Alps are slightly over-represented with an excess of about 120 ha ( Supplementary Table 1 ). It appears that the biogeographical pattern of the deficit in surfaces occurs mostly in the Alps ( Figure 1 ), but when looking at the finest strata level, lowlands in South-Western Switzerland should also be complemented to tend to a more balanced distribution across the strata ( Supplementary Figure S3 ). This analysis also reveals that Engadin, Eastern Plateau and the Rhine basin are already well covered by the existing network of surfaces, with only 15 to 50 ha to add in these regions ( Supplementary Table 1 , Supplementary Figure S2 ). The current distribution of the surfaces is correlated to the distribution of a proportional sampling, i.e. to the distribution of the potential candidate meadows (Pearson’s correlation = 0.81, p-value <0.001) but remains significantly different (Chi-square test p-values <0.001, Supplementary Table 1 ).

Next, based on floristic data controls, we assessed the actual proportions of each plant association in the instrument after two years of inception ( Figure 2 ). To check whether the frequencies of the associations that have been included in the in situ program are representative, we compared the current observed distributions with the modelled distributions of each association. The associations’ frequencies of the current surfaces are positively correlated with the associations’ potential distribution frequencies (Pearson’s correlation = 0.64) but this correlation is not significant (p-value = 0.086, Figure 2 ). A Chi-square supports that the current frequency of the associations is independent of the frequency of the modelled distributions of the associations (p-value = 0.016). This indicates a relatively even representativity of each association except for the overrepresentation of Poo pratensis-lolietum perennis and the underrepresentation of Cynosurion and Poion alpinae ( Figure 2 ). While Poion alpinae might not be the most threatened association considered in this program, its floristic quality is highly dependent on nitrogen levels/intensification and therefore very unevenly distributed (Eggenberg et al., 2015). It is also possible that many Poion alpinae surfaces are already integrated into other instruments dedicated to pastures, and therefore not considered here.

To refine the evaluation of the effectiveness of the in situ program, and to allow a diagnostic on whether and how to fill the rest of the surface quotas, we evaluate for each association their potential suitability and compare it to the actual surfaces already in the instrument. Visually comparing the suitability of each association to the actual extent of surfaces in the program allows a quick and easy diagnostic ( Figure 3 ). Typically, while a vast majority of the Arrhenatherion surfaces are located in the Jura and Eastern Switzerland, the suitability map flags other potentially interesting surfaces in the Western Plateau and the Western Central Alps ( Figure 3 ). Arrhenatherion representation in the program is almost comprehensive compared to the frequency of other associations ( Figure 2 ). However, more sampling effort is necessary in these underrepresented regions to cover the diversity across the full elevational and biogeographical gradient.

Primary targets. All 18 target species are logically distributed across the 1’217 current sites included in the in situ program. However, observation data were missing for 5 sites in the analysed dataset. The rarest species is Onobrychis viciifolia (9 sites), whereas Trifolium repens is the most widely distributed, observed in 1’029 sites ( Table 3 ). On average, each target species can be found in 391.17 +/- 326.86 sites. Each site includes between 1 and 15 target species, with an average of 5.79+/- 2.12 target species.

Other Priority CWR. The presence of at least one of the 18 target species is mandatory to qualify a surface for the in situ program. Except for Onobrychis viciifolia, these species are listed as national priority CWR. Meanwhile, the extent to which other priority CWR from the 285 taxa of the Swiss priority CWR checklist (Petitpierre et al., 2023) could also be present in these surfaces remains unexplored. These “secondary” targets could also be considered contributing to increase the actual effectiveness of this instrument. Other Swiss priority CWR were found in 358 sites (29%). These sites comprise between 1 and 18 supplementary priority CWR. In total, the current distribution of the sites includes 108 priority CWR (38% of the taxa; Supplementary Table 3 ). There are between 1 and 24 priority CWR per site, with an average of 6.76 +/- 2.88 priority CWR species per site. 9 sites (0.74%) comprise 8 species with a threatened Red List status (i.e. CR, DD, EN, NT, RE or VU). The contribution of the in situ surfaces to the overall diversity conservation appears therefore positive, not only for the primary targets but also for other priority CWR.

Most CWR conservation strategies focus on the “taxa” level (Maxted et al., 2006), but conserving biodiversity is more complex. Taking advantage of the example of in situ conservation of forage plants presented here, we would like to emphasize the various aspects that need to be considered for an efficient conservation design, namely specific and ecosystem diversity (taxa/species numbers and plant populations), phylogenetic diversity (particularly between crops and CWR) and finally intra-specific or allelic diversity.

The global governance of plant genetic resources historically first concentrated on ex situ conservation (Curry, 2022). Following a recent paradigm shift, it is now also considering in situ conservation as a critical approach to conserving biodiversity. In its second report on the State of the World plant genetic resources, the FAO wrote: “There is a need for more effective policies, legislation and regulations governing the on-farm management of plant genetic resources for food and agriculture, both inside and outside protected areas” (FAO, 2010). This view is combined with the Seed Treaty provision on the conservation of CWR in its Art. 5d: “Promote in situ conservation of wild crop relatives and wild plants for food production, including in protected areas, by supporting, inter alia, the efforts of indigenous and local communities” (FAO, 2010) and provide strong legal support to CWR in situ conservation. Several examples of in situ measures have already been reported, for example, the four types of in situ CWR measures spreading across more than 57 initiatives across Europe: CWR genetic networks, potential genetic networks, people and institutional networks and project-associated initiatives (Alvarez Muniz et al., 2020). It also appears that a vast portion of so-called in situ conservation happens passively, i.e. including CWR already included in an existing protected area. Surveys on in situ measures for CWR reveal a large array of practices and tend to overestimate the actual measures dedicated to CWR conservation (Rubio Teso et al., 2021). The CWR conservation, particularly for taxa that are not necessarily threatened or are situated on agricultural land (Petitpierre et al., 2023), does fall in a “grey zone” between conventional conservation (red-list based) strategies and ex situ collections for breeding purposes. The in situ program is an attempt to address this conservation gap.

One of the first challenges to answer to the requirements of the international framework is to build measures that would be adapted to the Swiss context, particularly to its complex federal structure and subsidies regime. After an extensive consultation and testing phase, the current model inception combined a top-down resource allocation covering up to a theoretical maximum 2’750 ha and a bottom-up approach, where farmers could decide whether to participate in the program. Now two seasons into the program, half of the surface’s quota has been assigned. Two major issues appeared: 1. The addition of a new -supplementary- instrument on top of the whole subsidy machinery is a clear disincentive for farmers. While the instrument concentrates on new surfaces not yet considered by other measures, it remains that the cost-benefit for participation might be perceived as too high. Integration of the in situ program into other existing subsidies or cross-compliance programs might be a possible alternative. Overall, participation in the in situ program was strongly correlated with the support and dedication of the local authorities. 2. The bottom-up approach does not allow for fine control over the surfaces to be targeted: it is not necessarily the best or enough surfaces that will be integrated at first (like the Poion alpinae deficit observed in the current state of deployment, Figure 2 ). A complementary approach that would specifically target complementary high-value populations could be a possible way forward. Some additional target associations could also possibly extend the scope of the instrument, like for example the acidophilic Festuca-Agrostis meadows. In any case, the involvement of local stakeholders and experts in the flagging of high-value surfaces would be necessary. However, favouring certain farmers over the quality of their surfaces could breach some legal requirements related to equal treatment.

The in situ program represents about 1% of the total amount allocated to biodiversity subsidies to agrobiodiversity and a very small fraction of the surfaces integrated in these schemes (representing in total 190’609 ha in 2022, (FOAG, 2022). Meanwhile, extensive meadows and pastures still represent the vast majority targeted by ECAs. Therefore, it may be important to consider the overlaps between in situ surfaces and other ECA surfaces in terms of plant associations, (given that one condition to add a surface to the in situ program is that they should not be covered by other ECAs) it remains possible that some species or populations would be already under the scope of other measures, for example as Q1 or Q2 surfaces. Identifying such overlap is essential to avoid redundancy and guarantee overall policy coherence. In addition, as with any wild species, the CWR taxa considered in the in situ program are generally also present outside the agricultural area, and the distribution and diversity of the population outside cultivated meadows should also be taken into account for conservation. In such cases, species-rich surfaces -may- be protected under the various networks of natural reserves and other protected areas. Meanwhile, 22% of Swiss priority CWR were not significantly better covered by protected areas (excluding the in situ surfaces) than any random species (Petitpierre et al., 2023). Taken together, the exact relationship between in situ surfaces as a relatively restricted portion of the overall grass (cultivated or natural) meadows emphasizes the complexity of evaluating the instrument, particularly at the taxa level.

Although it appears in our analysis that considering conservation of forage CWR at the population level can in practice be a promising way to mobilize resources at the farm level. An additional concern towards the efficiency of the instrument that cannot be disregarded is the timing of obligations. A minimal 8-year engagement was foreseen in the in situ program, with controls being held regularly. While 8 years is a significant time for a farmer, it remains a possible weakness that can hardly be addressed: It is hoped that once the first round is over, the political and farmer commitments will not be eroded. Long-term management (particularly along altitudinal gradients) should take into consideration climate change. Additional work is needed to evaluate the extent to which the in situ surface network will be impacted, and whether measures could be already taken to mitigate these changes. Several modelling have already been performed on local case studies, like in Holland (Treuren et al., 2017) and Norway (Phillips et al., 2017) and include recommendations towards better coordination with ex situ conservation.