An increase in food production in Europe could dramatically affect farmland biodiversity

Study regions and farms

Ten European regions from boreal to Mediterranean were selected (Supplementary Table 1). They represented major agricultural land uses such as arable crops including horticulture, mixed farming, grassland and perennial crops (vineyards and olives). Within each region, a pool of ~20–40 farms was selected from which 12–20 farms were randomly selected (169 in total) that belonged to the same farm type, produced under homogeneous climatic and environmental circumstances and fulfilled specific criteria regarding their main production branch. In case the selected farms were not willing to participate, we asked other farms from the pool till the sufficient number has been reached. The selected organic farms had all been certified for at least five years. Farmers were asked if they were willing to participate in the study. If they refused, additional random sampling was conducted. In the region NL, 11 organic farms agreed to participate but only three non-organic farms, whereas seven organic farms and 11 non-organic farms were available in the region HU. During the study, one non-organic farmer in the region CH ceased participation.

Habitat maps and farm interviews

The complete area of all selected farms was mapped, using the BioHab method36. Excluded from the farm area were woody and aquatic habitats larger than 800 m2 and summer pastures. Within the farm area, areal and linear habitats were recorded. For an areal habitat, the minimal mapping unit was 400 m2 with a width of at least 5 m. More narrow habitats, between 0.5 and 5 m wide and at least 30 m long, were mapped as linear habitats. Habitats were distinguished in habitat types according to Raunkiær life forms, environmental conditions and management evidence28. Further, a farmland class was assigned to each habitat that described whether the habitat was managed for agricultural production or other objectives such as e.g. nature conservation. In face-to-face interviews following a standardized questionnaire, farmers provided detailed information on field management and yield.

Categorization as production fields and semi-natural habitats

Based on the habitat maps and available information about management intensity, we categorized all habitats as either semi-natural habitats or production fields. In agricultural landscapes, these two categories are often not clearly distinguishable. There is a gradient from more intensively managed production fields to less intensively used semi-natural habitats. In addition, a categorization at the local scale can be different from an approach at a European scale (29 and see p. 45 of37). Here, we applied the same criteria for all ten study regions.

In all cases, we categorized as production fields: arable crops, intensively managed grasslands (following main plant species observed, management evidence and objectives, with fertilization and/or two or more cuts a year), horticultural crops, and vineyards.

We categorized as semi-natural habitats: linear habitats, habitats that were managed for nature conservation objectives, habitats where mainly geophytes, helophytes or hydrophytes were growing, grasslands with woody vegetation (shrubs and/or trees), and extensively managed grasslands (no fertilization, no or one cut a year).

Species sampling

Vascular plant, earthworm, spider and bee species were sampled in all different habitat types of a farm. One plot per habitat type was randomly selected per farm for species sampling. This resulted in 1402 selected habitat plots on 169 farms (Supplementary Table 2). In the selected habitats, species were sampled during one growing season, using standardized protocols19,38. Plant species were identified in squares of 10 × 10 m in areal habitats and in rectangular strips of 1 × 10 m in linear habitats. Earthworms were collected at three random locations of 30 × 30 cm per habitat. First, a solution of allyl isothiocyanate (AITC) was poured out to extract earthworms from the soil. Afterwards, a 20-cm-deep soil core from the same location was hand sorted to find additional specimens. Identification took place in the lab. Spiders were sampled on three dates at five random locations per habitat within a circle of 0.1 m2. Using a modified vacuum shredder, spiders were taken from the soil surface, transferred to a cool box, frozen, or put in ethanol, sorted and identified in the lab. Bees (wild bees and bumble bees) were sampled on three dates, during dry, sunny and warm weather conditions. They were captured with an entomological aerial net along a 100 m long and 2 m wide transect, transferred to a killing jar and identified in the lab.

Grouping of species data

Species data were pooled per taxa, habitat and region, and three sub-communities were formed with species (1) exclusively found in semi-natural habitats, i.e. unique to semi-natural habitats, (2) exclusively found in production fields, i.e. unique to production fields, and (3) found in both habitat categories i.e. shared by production fields and semi-natural habitats. For calculations of effects over all four taxa, species richness was the sum of the individual taxa species richnesses.

Estimating species richness

Species richness was estimated using coverage- and sample-size-based rarefaction and extrapolation curves31,39,40. Rarefaction and extrapolation, including confidence intervals (bootstrap method) and sampling coverage, were calculated in R 3.4.041 using package iNEXT42. Detailed information is provided below for each topic.

Estimating richness of unique species to compare semi-natural habitats and production fields

To legitimately compare the richness of species unique to semi-natural habitats and to production fields, we used the coverage-based method, i.e. we standardized the samples by their completeness30. The point of comparison was determined by the so-called ‘base coverage’ identified by the following procedure31: (1) select the maximum sample coverage at reference sample size (number of sampling units) of the sub-communities under comparison, (2) select the minimum sample coverage at twice the reference sample size of the sub-communities under comparison, (3) identify the maximum of the results from step (1) and step (2) as ‘base coverage’. The species richness estimates were then read off from the species sample-size-based rarefaction and extrapolation curves at the ‘base coverage’ for each sub-community being compared. If zero or exactly one species was unique to a sub-community at the reference sample size, no sample coverage could be calculated. In this case, we set the species richness at 0 or 1, respectively. The species richness estimate of the other sub-community under comparison was then read off at twice the reference sample size on the curve.

The ‘base coverage’ was individually defined for each region and each taxonomic group since the mixed effects models used to analyze the data took into account the variation among regions and taxonomic groups.

Differences in species richness unique to semi-natural habitats and production fields

The difference between the species richness unique to semi-natural habitats and unique to production fields was tested with mixed effects models using package lme4 (Version 1.1-12) in R43. The data were (Sij| β, b, x) ~ Poisson(µij) from i = 1, …, 10 regions. The model is:

$${{{rm{ln}}}}left({mu }_{{ij}}right)={beta }_{0}+{beta }_{1}{x}_{1i}+{b}_{1i}$$

(1)

$${b}_{1} sim N(0,sigma 2)$$

where ({beta }_{0}) is a fixed intercept, ({beta }_{1}) a fixed effect sub-community ({x}_{1{ij}}) (species unique to semi-natural habitats versus species unique to production fields), b1i are random intercepts for region i. Random effects are normally distributed with mean 0 and variance σ2. The significance of term ({beta }_{1}) was calculated by log-likelihood ratio tests with one degree of freedom. For the models over all four taxa, an additional random intercept was included, i.e. b2j with mean 0 and variance σ2 for j = 1, …, 4 taxa (Fig. 1b).

Differences in species richness between organic and non-organic systems

The comparison between organic and non-organic systems of species unique to semi-natural habitats and to production fields, and of species shared by the two habitat categories, relied on coverage-based extrapolation as described above. Differences between management systems were tested for significance using mixed-effects models with management system ({beta }_{1}) ({x}_{1{ij}}) as fixed effect in (1).

Estimating species loss due to conversion of semi-natural habitats to production fields

To predict the species loss due to conversion of semi-natural habitats to production fields, we relied on sample-size-based extrapolations31 with species incidence frequencies. We estimated the richness of the species pool for the total number of mapped habitats including the extrapolated species richness unique to semi-natural habitats and unique to production fields, and the observed richness of shared species for each of the four taxa. This species pool provided the basis for the calculation of the species loss or gain (Table 1 and Supplementary Table 7). To model the species richness decrease for any amount of semi-natural habitats converted to production fields, we calculated and drew backward the curve composed of the accumulation curve for species unique to semi-natural habitats, to which the estimated total species richness unique to production fields (constant) and the corresponding gain of species unique to production fields (increases with increasing area of production fields as semi-natural habitats are converted), and the richness of observed shared species (constant) were added. This is the species decrease curve (Supplementary Fig. 2). If started at the observed species richness, this curve corresponds exactly to a species richness curve calculated by a cumulative random removal of semi-natural habitats one by one from the pool of all habitats. The four taxa decrease curves were added for the curve in Fig. 2. Confidence intervals (CI, 95%) shown in Figs. 2 and 3 are calculated by bootstrapping within the calculation of the species accumulation curves (iNEXT42), upper and lower bounds of the 95% CI of the four taxa being added. From the species decrease curve, we read off the predicted species richness for a conversion of 50% and 90% of the semi-natural habitats, and a conversion required to increase production by 10%.

As species were sampled in 20% of all mapped habitats on average per region (min. 8%, max. 35%), extrapolated species accumulation curves used to build the species decrease curve were calculated for more than two to three times the reference sample size, which is the suggested range for reliable extrapolation of the species richness estimator31,44. Obviously, the confidence intervals (CI) of the species richness extrapolations here became wide (Supplementary Fig. 4). As we still wanted to show the impact of a conversion of the whole semi-natural area into production fields on the production gain in the ten regions, we used the uncertainty (upper and lower bounds of the 95% CI of the four taxa added) to define two situations in addition to the average case to predict species richness for a 50% and a 90% semi-natural habitat conversion, and a conversion required to increase production by 10%: (1) a worst case situation with the upper bound of the CI of the expected species richness unique to semi-natural habitats, the lower bound of the CI of the expected species richness unique to production fields, and shared species assumed not to be able to survive without semi-natural habitats and considered like species unique to semi-natural habitats (i.e. upper bound); and (2) a best case situation with the lower bound of the CI of the expected species richness unique to semi-natural habitats, the upper bound of the CI of the expected species richness unique to production fields, and the lower bound of the CI of the expected shared species richness.

Estimating production gain

Farmer interviews delivered an average yield per crop type per farm for the years 2008–2010 (Supplementary Data45 shows details for organic and non-organic systems separately). Farmers indicated yield in kilograms or tons per hectare. This was transformed into energy units, i.e. mega joules per hectare (MJ ha−1) using standard values46. From this, for each region, the average yield (MJ ha−1) was calculated by first multiplying individual crop type yields by the corresponding crop type areas to obtain the production per crop type, then summing up the production of all crop types, and finally dividing this sum by the total area of the crop types. For livestock farms, the fodder production of grasslands was estimated based on the average requirements per livestock unit, accounting for the amount of feed grain, legumes, silage maize and of imported feedstuff. All yields relate to plant biomass production and do not comprise livestock products. The average yield takes into account the relative cover of the different crop types in the regions. Therefore, the conversion of the semi-natural area to production fields was region-specific. The production of certain semi-natural habitats as e.g. olive groves in Spain was not part of the production calculation. The reason is that data on production for semi-natural habitats were mainly not available and/or negligible, e.g. extensively used grassland in CH or in HU, and we decided to apply the same treatment to all the regions. Consequently, in case of olive groves in Spain the effective increase in production is overestimated. To calculate the production gain per region, the production field area added by the conversion of semi-natural habitat area was multiplied by the average yield. In practice, in many regions it may be impossible to convert semi-natural habitat to productive land due to geomorphological constraints and poor soils, and even if land were converted, yields would be much lower than these averages. The results presented here, especially the 90% scenario, are therefore over-optimistic. On the other hand, our calculations are based on the area of semi-natural habitat available for conversion on existing farms, but in some regions other sources of semi-natural land may be available for conversion, e.g. former agricultural land that has been abandoned.

Species loss and production gain for three scenarios

We calculated the change of species richness and the production gain under current day production efficiency for two scenarios: (1) a conversion of 90% of the semi-natural area into production fields. The 10% of semi-natural area remaining is considered unsuitable for agricultural use or even impossible to cultivate; (2) a conversion of 50% of the semi-natural area into production fields, and (3) a necessary conversion of the semi-natural area into production fields to achieve a 10% production increase per region.

Standardization for organic and non-organic systems

Although the overall mapped area, the number of semi-natural habitats, the number of production fields and the average habitat size did not significantly differ between the two management systems (Supplementary Table 5), we standardized the number and size of habitats to the average across both systems per region to compare the species loss and production gain at current day production efficiency in the organic and non-organic systems. The total production in organic and non-organic systems per region was calculated based on the respective yield and the average mapped area of the production fields across both systems as described in section “Estimation of production gain”. The impact on biodiversity was analyzed for the scenario that organic systems should achieve the same level of production as non-organic systems by converting semi-natural habitats to production fields. We calculated the amount of the required area to be converted into production fields and the corresponding species change.

Differences between management systems were again tested for significance using mixed-effects models with management system ({{{{rm{beta }}}}}_{1}) ({{{{rm{x}}}}}_{1{{{rm{ij}}}}}) as fixed effect in (1).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

admin

Next Post

Real-life ‘Jetsons’? Watch robots cheerfully serve food at this Dallas restaurant

Wed Sep 8 , 2021
La Duni’s newest staff members in Dallas are quick on their feet and quite polite. One even said, “I love you, you are so pretty,” after a delivery. Meet Alexcita, Panchita and Coqueta, La Duni’s restaurant robots. All of La Duni’s robots have personalities. Customers “don’t really see them as […]