Fertilization using manure minimizes the trade-offs between biodiversity and forage production in agri-environment scheme grasslands

A common practice used to restore and maintain biodiversity in grasslands is to stop or decrease the use of fertilizers as they are a major cause of biodiversity loss. This practice is problematic for farmers who need fertilizers to increase forage and meet the nutritional needs of livestock. Evidence is needed that helps identify optimal fertilizer regimes that could benefit biodiversity and livestock production simultaneously over the long-term. Here, we evaluated the impact of different fertilizer regimes on indicators related to both biodiversity (plant, pollinator, leaf miners and parasitoid Shannon-Weiner diversity, bumblebee abundance, nectar productivity and forb species richness), and forage production (ash, crude protein, ruminant metabolizable energy and dry matter). To this end, we used data from a grassland restoration experiment managed under four nutrient inputs schemes for 27 years: farmyard manure (FYM; 72 kg N ha-1 yr-1), artificial nitrogen-phosphorus and potassium (NPK; 25 kg N ha-1 yr-1), FYM + NPK (97 kg N ha-1 yr-1) and no-fertilizer. Results showed strong trade-offs between biodiversity and forage production under all treatments even in applications lower than the critical load in the EU. Overall, farmyard manure was the fertilizer that optimized production and biodiversity while 97 kg N ha-1 yr-1 of fertilizer addition (FYM+NPK) had the most negative impact on biodiversity. Finally, forage from places where no fertilizer has been added for 27 years did not meet the nutritional requirements of cattle, but it did for sheep. Rethinking typical approaches of nutrient addition could lead to land management solutions suitable for biological conservation and agriculture.


Section 1 Colt Park trail experiment
Colt Park trail is a long-term nutrient and plant biodiversity manipulation experiment located at 300 m altitude in the Ingleborough National Nature Reserve in North Yorkshire, England (54°12'N, 2°21'W).The field trial started in 1990 on permanent grassland dominated by the perennial grass species Lolium perenne and Cynosurus cristatus, on a shallow brown-earth soil (pH 5.1) over limestone of moderate-high residual fertility.The aim of the experiment was to test different management strategies for improving the diversity of grasslands in a working agricultural context (1,2).
The experiment consists of 72 plots, each 2.5 m x 6 m (15 m2) in size, arranged in three blocks of 24 plots.Each of the three blocks is subdivided into three sub-blocks of 8 plots corresponding to three sowing treatments applied during 2004-2008 with no remnant effect on the biological communities studied (3).For this, experiment we used only the plots under no seed addition treatment, this selection results in 8 plots per block for a total of 24 plots.In each block three fertilizer treatments have been applied to two plots randomly chosen plots: N:P: K fertilizer (20:10:10, 25 kg ha -1 nitrogen plus 12.5 kg ha -1 of P2O2 and K2O; hereafter NPK,), farmyard manure (12 t ha -1 hereafter FYM), N:P: K fertilizer + farmyard manure (hereafter NPK+FYM) and a further two plots are left as a control with no fertilizer.In total, there are 6 plots per treatment.
The quantities are according the limits of the environmental stewardship, of 12,500 kg (total rate of nitrogen must not exceed the 100 kg/ha), per hectare per year for FYM and 50kg for inorganic fertilizer (4) versus 91 kg/ha of inorganic fertilizer outside stewardship agreements, rising to well over 200 kg for multiple harvest silage systems (5).

Section 2 Estimation of livestock production
To estimate the number of animals that could be fed with the hay produced under each treatment, we first estimated the amount of hay that would be available for daily animal consumption under each treatment.To this end we divided the amount of hay that would be available for animal consumption everyday by the amount of hay that an animal would need to consume daily to reach its nutritional needs.To this end, we first transformed the mean grams per meter of hay produced and scaled up to one hectare.Then, we calculated the amount of hay that could be consumed in a day by the cattle.This was done to ensure that the hay produced lasted the standard 24-week period that the cattle are reared indoors during the winter (16).The latter was estimated by dividing the amount of hay produced in one hectare by the total number of days the cattle are indoors (168 days).Finally, we divided the amount of hay by the amount an animal would need to consume considering the metabolizable energy of the hay using the app FarmIQ (2016).In each cell v denotes the vector of ranked measurements across the 12 indicators min-max normalized.Two test statistics were generated for the observed vectors: the difference in weighted mean and weighted standard deviation of the indicators in each vector between each pair of treatments.The same test statistics were calculated for each random reshuffle of the data.Hence vectors were swapped around within each block but with the randomization carried out across all blocks simultaneously.Test statistics were calculated for each randomization, and these form a reference distribution.A P value conditional on the reference distribution was then calculated as the proportion of times the absolute value of each observed test statistic was greater or equal to each randomized test statistic.Significant P values would fall in the tails of the reference distribution leading us to reject the null hypothesis of no difference between the observed and randomized test statistics for the pair of treatments.to unequal number of variables, we compared the weighted mean and the weighted standard deviation.We defined as optimization as the point with the highest weighted mean but the lowest weighted standard deviation.In this sense the optimal treatment is that with the highest weighted mean and lowest variation around the mean (measured as the weighted standard deviation), represented with a circle in the figure.Table 3. Probability that first treatment (T1) is higher than the second treatment (T2; P(T1 >T2)) and probability that T1 is lower than T2 (P(T1< T2)) for each of the pair comparisons for the mean across variables within each group.The p values correspond to the of times out of 10,000 randomizations the mean of the treatment T1 was lower/higher than the mean of the treatment T2.SIG stands for significant (P <0.05).4. The probability of each treatment of maximizing the mean across variables and minimizing the variance around the mean.The probability value corresponds to the number of times out of the 10,000 randomizations the treatment on the left has a higher value than the treatment on the top.We considered probabilities values to be significant lower when P = < 0.05 and significant higher when P => 0.95.Those significant values are indicated with *.

Fig 1 .
Fig 1.The randomization testing scheme is shown for one of the three experimental blocks and for one pair of treatments only.In each cell v denotes the vector of ranked measurements across the 12 indicators min-max normalized.Two test statistics were generated for the observed vectors: the difference in weighted mean and weighted standard deviation of the indicators in each vector between each pair of treatments.The same test statistics were calculated for each random reshuffle of the data.Hence vectors were swapped around within

Fig 2 .
Fig 2. Optimizing values across variables for each treatment.The x-axis represents the mean across variables while y-axis represents the variation or standard deviation around the mean.Due

Table 2 . Performance of each variable for each fertilizer treatment.
Values were min-max normalized so values can be compared between treatments and across variables.The mean value for each group of variables is reported as well as the weighted mean across variables.We weighted the value by considering that variables of forage should count for the 50% of the mean value and variables insect community and plant community will contribute for the other 50%.Neutral cellulase gammanase digestibility was removed from the analysis due to high correlation with ruminant metabolizable energy.Standard errors are reported next to mean values.

Table 5 .
Livestock production under fertilizer treatment.Forage production and feeding value relative to winter requirements of spring-calving suckler cows and April-lambing mixed age ewes.