Introduction

High livestock densities can cause regional accumulations of slurry and farmyard manure, leading to drastic nitrogen (N) and phosphorus (P) surpluses in agricultural soils in the region (Svanbäck et al. 2019). Transporting this organic fertilizer to distant areas with a need for these nutrients is not lucrative due to the high water content, especially considering current soaring energy costs (European Commission 2022a, b). Even in areas with high livestock densities, additional mineral fertilizers are often used to fulfill plant needs, as the nutrient ratio in manure is suboptimal. Mineral fertilizers have a poor climate balance and a high price volatility for farmers (Hasler 2017; Schnitkey et al. 2023). Therefore, regional production of mineral fertilizers from existing material flows such as biomass, manure, and digestate could help reduce imported fertilizers and make regional farming systems more resilient. The European Innovation Partnership (EIP) “Agriplus Hohenlohe” (EIP‐AGRI 212018) is an example of a case study that addresses this issue in the Hohenlohe region (south-west Germany) and aims to increase the efficiency of arable farming through improved nutrient management. By building a nutrient recycling plant on an industrial scale, manure from local farmers is processed into mineral fertilizers after being used for biogas production (European Commission 2023). These fertilizers can be used specifically in arable farming to close nutrient cycles. One of the products of the nutrient recycling plant is the mineral fertilizer liquid ammonium sulfate (LAS).

Studies with LAS were conducted in the form of pot/greenhouse (Sigurnjak et al. 2016, 2019) and on-station field experiments (Vaneeckhaute et al. 2013b, 2014; Mokry 2013; Sigurnjak et al. 2016, 2019). All studies found the fertilizing performance of LAS to be satisfactory. Even though it is important to test non-conventional fertilizers in real agricultural systems for their effect on yield (Sigurnjak et al. 2016), to our knowledge, no on-farm field experiments have yet been performed. The owner of the nutrient recycling plant and some neighboring farmers as potential users of the products asked the University of Hohenheim to test LAS and suggest ways to implement it into their farming practices. In initial meetings, it became clear that the trials should take place on the farmers' own land and that this should be done as a participatory investigation to target the outcomes of the study specifically to the farmers’ needs. Studies conducted with a participatory approach have found that a close collaboration between farmers and scientists supports innovations in agriculture, increases innovation acceptability when farmers are in control of the experimental process, and produces more relevant technologies and greater economic impacts, especially when participation starts early in the research process (Bellon et al. 2002; Johnson et al. 2003; Padel et al 2015).

Mokry (2013) conducted an on-station field experiment from 2009 to 2012 in the same region and indicated that a star wheel application technique is optimal for LAS. However, the fact that farmers can only use this technology for LAS makes it too expensive to acquire. They would rather use their existing technology. In addition, the star wheel technique takes a relatively long time to apply fertilizer, which increases the variable costs for the farmers.

The objectives of this study were to implement knowledge from on-station field experiments in on-farm best practice, to compare established fertilizing strategies with recycled LAS, and to investigate whether recycled LAS could substitute the commonly used mineral fertilizer calcium ammonium nitrate (CAN) in the future. The following research questions were derived:

  1. 1.

    How can scientific methods be successfully applied to a fertilizer experiment in an on-farm setting and scale?

  2. 2.

    What is the most suitable application technology for LAS in farming practice?

  3. 3.

    Can the recycling fertilizer LAS serve as a substitute for CAN in common fertilizing regimes?

Materials and methods

Location and climate

A two-year (2020–2021) field experiment was conducted in the Hohenlohe district in the Heilbronn-Franken region at an altitude of 300 to 348 m. Compared to the average for the federal state of Baden Württemberg, the proportion of the population employed in agriculture in this region is high. Due to the diversity of agricultural products, the Hohenlohe district is considered a good reflection of the entire agricultural spectrum of the federal state (except for hops and tobacco). The agricultural land area ranges from gently undulating, very fertile plateaus to steep shell limestone valleys of the main rivers and their tributaries (Hohenlohe municipality administrative office 2023). There are two weather stations (the German Weather Service “DWD” and the Agricultural Technology Centre Augustenberg “LTZ”) located at a distance of less than 10 km from the experiment site. The DWD also provides long-term weather data. The climate data from this station for 1981—2010 gave a mean annual average temperature of 9.8 °C (2 m), annual precipitation of 841 mm, and 1735 h annual sunshine. In 2020 (LTZ), a mean temperature of 11.4 °C (2 m), annual precipitation of 576 mm, and an annual total of 1712 h sunshine were recorded. In 2021, the mean temperature was 9.8 °C (2 m), the annual precipitation 646 mm, and the annual sunshine 1509 h. The monthly precipitation and long-term average for both experimental years are shown in Table S1. A late spring frost occurred in the morning hours of 12 May 2020 when winter barley (Hordeum vulgare L.) was in BBCH 60-65 (flowering). This frost caused damage to the flowering barley stands throughout the region (Strotmann 2020; LFL 2020).

Fields and crops

The farmers who agreed to participate in scientific trials all run livestock farms with less than 100 ha land and were interested in further closing nutrient cycles on their farms. The experimental fields selected with the farmers for this two-year experiment were intended to represent conditions typical for the region. Two fields from each of four local farmers (F1, F2, F3, F4) were selected and are denoted “a” and “b” in this paper, e.g., “F3a” and “F3b”. Seven plots were established in each of the eight fields. The width of the plots was 15 m (Table 1), which corresponded to the working width of the drag hose spreader for slurry and fermentation product as well as the field sprayer available on the farms involved. Due to the topography of the fields, each plot had a different length (83–286 m) which resulted in different sizes for each plot (Table 1 and Figure S1a, b). To characterize the field sites, composite samples were taken in autumn 2019. The texture of all plots was determined from composite soil samples for the analysis of mineral nitrogen (Nmin) in February 2020. All fields in this experiment are classified as stagnic Luvisols (FAO 2015) with an average of 27% clay, 8% sand, and 65% silt. Farmers F3 and F2 raise piglets, F1 has fattening pigs, and F4 has dairy cattle. The average pH of all fields was 6.72 in 2020 (min. 6.25, max. 7.3). The position of the treatment in each field remained constant across years (Figure S1a, b). Crop rotations were established in all fields; thus, in each field, the crops differed between years (Table 1). Winter barley and winter wheat (Triticum aestivum L.) were sown at a row spacing of 150 mm in 2020. Winter rapeseed (Brassica napus L.) was sown in autumn 2020, and maize (Zea mays L.) (row spacing 750 mm) was sown in May 2021 after a legume-free intercrop.

Table 1 Fields of the farmers F1, F2, F3, and F4 including field and plot sizes, crop rotation, and N + P concentrations in the slurry (original substance)
Full size table

Participatory approach

The objective of the participatory approach was to co-design and implement an experimental setting for an on-farm trial to enable the elaboration of practical solutions for the efficient application of recycled LAS. The goal was to make use of the local farmers’ knowledge as much as possible as a key to the success of the on-farm experiment.

The participatory approach was applied in two phases:

  1. 1.

    Co-design of the on-farm experimental design

On-farm experiments can fail due to a lack of compromise between scientists and practitioners, logistical support and farmer participation, if they are based on formal, complex, researcher-designed set-ups (Lightfoot and Barker 1988). Such problems in on-farm experiments can stem from superficial communication and interaction between researchers and farmers (Hoffmann et al. 2007). In addition to unbiased results, curiosity, and the potential for increased profit, a large motivating factor for farmers is to act as equal partners in the project (Thompson et al. 2019). Therefore, we decided to maintain close communication and social contact with the farmers in a participatory approach.

In addition to many bilateral conversations with farmers during field visits, sampling and telephone calls, regular official meetings took place with the farmer group. These meetings were held every 3 months beginning 14 months before establishment of the on-farm field design and start of the experiment. The meetings were held on the farm premises of the farmer who established the nutrient recovery plant. All four farmers and the nutrient recovery plant owner involved in the field trial participated in the meetings. They were held in a workshop format where scientists prepared input with ideas on the design and management of the on-farm field trial and discussed them with the farmers. In each workshop, the results of the previous meetings were summarized in a presentation as a basis for the following decisions. The workshops also included excursions to the farmers' stables and fields to define the framework conditions.

The broad design of the trial was devised within 10 months so that the winter cereals could be sown before the trial started with the first fertilization in spring 2020. For both the scientists and farmers, it was important to establish a friendly basis for cooperation at eye level, which established an "all in one boat" mindset.

In the workshops, farmers identified their own existing technologies, such as a field sprayer with three-jet nozzles in the first year and the slurry tank with a dragging hose in the second year, as a substitute for the star wheel application technique. We agreed that operations such as soil cultivation, plant protection, and fertilization should always be carried out by the same farmer or his employee for the same crops in the experiment in order to treat all fields equally. Any kind of data collection, like plant and soil sampling, should be carried out by the scientists, who also had the task of coordinating the harvest and taking samples during it.

  1. 2.

    Cooperative performance of the on-farm trial

During the experimental phase, meetings were held before fertilization in 2020/21 to discuss the logistics of fertilization and plant protection and again after harvesting to discuss initial results. The regular measurements of chlorophyll content during crop growth gave ample opportunity for bilateral conversations with the farmers to discuss topics such as the management of harvest logistics. In this way, any problems were solved and questions answered directly face to face on the fields or via telephone calls. Spontaneous adaptations to circumstances requiring changes to the experimental plan developed in the meetings that affected all farmers were communicated to the group as a whole via a common smartphone app. Farmers and scientists worked together cooperatively, helping each other with practical needs like measurements and sampling (task of scientists) or machinery operations (task of farmers).

Experimental set-up

In the first meetings with the farmers, the decision was made to establish seven plots in each field (V1–V7) (Figure S1). Due to production problems of the nutrient recycling plant at a later stage of the project, only one fertilizing product (LAS) could be tested. As a result, in these seven plots, only five different fertilizing treatments (T1–T5) were established, but with two replicates per field for the recycled mineral fertilizer (LAS) and a common mineral reference (CAN) (Table 2). To further maximize the practicability and flexibility of this on-farm experiment, the seven plots were randomized as plots per field that cover the full length of the fields (Table 1), without repetitions in each field. The fields (n = 8) were treated like blocks in a randomized complete block design, since it is not mandatory to have repetitions in the same environment or field to conduct experiments comparable to on-station experiments in terms of precision (Schmidt et al. 2018; Piepho et al. 2011; Schulz et al. 2015).

Table 2 Description of fertilizer treatments and number of replications in each field
Full size table

As the fields were often slightly S-shaped, GPS software was not used and plots were established by hand. By creating tramlines in the middle of the 15-m-wide plots, it could be ensured that, for example, the field sprayer or the combine harvester is driven exactly along an S-shaped plot.

Fertilization strategy

The fertilizer requirements of the crops were based on the German fertilization legislation (DüV 2020) in order to operate with a practical fertilizer regime. In Germany, the N requirement of the crop is determined by the expected crop yield and the presumed N uptake of the crop, making an allowance for the non-harvestable residual plants. Then, field-specific factors such as autumn fertilization, humus content, past organic fertilization, catch crops, and mineral N content of the plots are subtracted. Since by legislation the maximum amounts of nitrogen and phosphorus in organic fertilizers are limited, triple super phosphate (TSP) and CAN were applied to balance the sum of total N and P in all treatments to make them comparable. All four farmers used their own manure to fertilize the manure-treated plots in their field. For this reason, the N and P contents (pig and cow slurry) were analyzed before fertilizing (Table 1). For the farmers F1, F2 and F3 the organically bound N fluctuated between 14 and 17% (pig husbandry) and for farmer F4 around 55% (dairy cows). The digestates contained 0.15% P and 0.47% N (42% of which organically bound). The LAS had 8% N and 9% S content.

In 2020, the winter wheat plots were fertilized with 198–209 kg N ha−1 split into three applications (tillering 19.-24.03.2020 (BBCH 21); shooting 25.-29.04.2020 (BBCH 30-34); flowering 08.06.2020 (BBCH 61-69)). The winter barley plots were fertilized with 109–168 kg N ha−1 split into two applications for tillering and shooting. LAS was applied with three-jet nozzles of a field sprayer, manure and digestates with a slurry tanker-trailing hose team, and the granulates of CAN and TSP with a fertilizer spreader. The total P applied for all crop treatments in 2020 and 2021 was 40 kg P ha−1.

In 2021, the slurry tanker-trailing hose team used for manure and digestate was also used for LAS, which was mixed with water to adapt the N concentration of the solution to the voluminous application rate of the conveyor technology (Figure S1d). CAN and TSP were applied using a fertilizer spreader. All plots of maize were fertilized with 70.5–87.5 kg N ha−1 before sowing at the end of April and then incorporated while preparing the seedbed. As a cruciferous plant, winter rapeseed is sulfur-sensitive. Therefore, all plots sown with winter rapeseed in autumn 2020 received a basal fertilization of 40 kg S ha−1 and 50 kg Mg ha−1 to compensate for possible effects of the sulfur contents of LAS and the organic fertilizers. Rapeseed plots were fertilized at the beginning of March and April with a total of 138–183 kg N ha−1.

Data collection during growth

Chlorophyll content assessment

To test for chlorophyll content in leaves, single photon avalanche photodiode (SPAD) measurements were taken using a SPAD meter (SPAD-502, Konica Minolta INC, Japan) for nine weeks in 2020 between the first N application and the beginning of dough maturity (BBCH 80). This resulted in three measurement dates for barley and five for wheat. Measurements were taken randomly 16 times per plot. Because changing light conditions influence the translocation of N in the leaves (Rousseaux et al. 1999; Dreccer 2000), measurements were only taken on non-cloudy days. To reduce further inaccuracies, measurements were taken until at least the tenth day after the last fertilization and only the youngest fully developed leaves were sampled.

Biomass collection

To be able to confirm and discuss the data from yields harvested with conventional combines, whole aboveground plant biomass was collected at dough maturity (BBCH 85) in 2020. In barley, this was done in the fields of farmer F4 on 09.06.2020 and F2 on 14.06.2020 after 245 days of growth. In wheat, the biomass cut was carried out in the fields of farmer F3 on 06.07.2020 after 267 days of growth and at F1 on 08.07.2020 after 252 days. Samples of 1 m2 were taken twice per plot. The fresh mass was collected and then the stems were separated from the spikes which were counted. Stems and spikes were dried separately at 60 °C for two days, the dry mass of stems and spikes was then determined (Sartorius F618 D2 balance, Sartorius AG, Germany).

Harvest in 2020 and 2021

Since this is an on-farm experiment, the harvest was done using the farmers’ own combines. First, the headland of the experimental areas was threshed, before threshing the rest of the plots. This allowed the technical setting of the combine to be adjusted for the harvest of the plots to ensure that each plot was treated equally by the machine. Directly after the harvest of the center of each plot (Figure S1f), the actual length of the plot was determined. Interfering areas, e.g., where wild boars had damaged the crops, were excluded from the respective total plot area of the treatments. The harvest of each plot was unloaded onto its own trailer (Figure S1c). During the unloading process, the samples were taken, and trailers were then weighed on a truck scale (60,000–400 kg, e = 20 kg). Up to six trailers shuttled between the experimental area and the scale during the harvest of each plot. A routing system with post-it notes on the truck driver's windshield was used to coordinate the trailers during harvesting.

Since it was not technically possible to determine the straw yield in 2020 with a baler at plot level, three samples were taken by hand from each plot. For this purpose, a two-meter-long folding ruler was randomly thrown into the respective straw swath (Figure S1e). The field-dried straw mass of the two meters was bagged and the composite sample was then weighed using a floor scale (1500–0.5 kg). From the cutting width of the respective combine harvester and the plot length, a straw yield per area in t/ha was thus obtained for each plot. Due to the weather conditions in 2021, it was not possible to weigh the fresh matter of all the ~ 1-kg maize samples (shredded) taken from the trailers fast enough without desiccation losses. For this reason, three samples (whole plants) were taken from the headland of each field to determine the water/dry matter content. To validate our methods of measuring grain and straw yield, we compared them with the results of the 2-m2 biomass cuts (weight, number of spikes, stem weight; 2020 only) using a bivariate model.

Soil samples and analysis

Mixed soil samples were taken from the fields in autumn 2019 for texture and CAL-P determination (0–0.3 m) and in spring 2020 for Nmin determination (0–0.9 m) using a Pürkhauer auger. Soil texture was determined by sieving and sedimentation using the ISO 11277 method. After the straw had been removed from the fields, plot-specific soil samples (0–0.3 m) were taken before tillage. These were air dried, sieved, and then analyzed for pH, CAL-P, and CAL-K.

To determine the Nmin content, the soil samples were extracted with CaCl2 and then measured using FIA (VDLUFA I A 6.1.4.1) (Schuller 1969). For the determination of plant-available P and K, the soil samples were extracted with calcium acetate lactate and measured using a flame photometer and FIA, respectively (VDLUFA I 6.2.1.1). The soil pH was determined by CaCl2 extraction with subsequent measurement using a glass electrode (DIN ISO 10390:2005) (VDLUFA 2012).

Statistical analysis

Data from each year had fixed and random effects, so they were analyzed using a mixed model approach. An analysis across years was not performed, as crops differed between the two years due to crop rotation. The following model was used:

$$y_{ijkl} = \mu + \tau_{i} + \varphi_{j} + \left( {\tau \varphi } \right)_{ij} + f_{k} + b_{kl} + e_{ijkl} ,$$
(1)

where \({y}_{ijkl}\) is the observation of treatment i in crop j on the lth field of farmer k, \(\mu\) is the intercept, \({\tau }_{i}\) is the fixed effect of treatment i, \({\varphi }_{j}\) is the fixed effect of crop j, \({(\tau \varphi )}_{ij}\) is the fixed interaction effect of treatment i and crop j, \({f}_{k}\) is the random effect of farmer k, \({b}_{kl}\) is the random effect of field l, and \({e}_{ijkl}\) is the compounded effect of plot effect and error of \({y}_{ijkl}\) with a crop-specific variance. Pre-requirements for analysis (normal distribution and homogeneous variances of residuals) were checked graphically via residual plots. Where deviations were found, data were logarithmically transformed prior to analysis. In these cases, means were back-transformed for presentation purposes only. Standard errors were back-transformed using the Delta method. Where significant differences were found for one term, corresponding means were compared using Fisher's LSD test. A letter display was used for description of multiple comparisons.

To estimate the correlation between two response variables, a bivariate model was used (Piepho and Möhring 2011). Model (1) was extended into three steps: First, fixed effects were replaced by random effects. Second, a factor trait with two levels for the two response variables was defined. Third, all effects in the modified version of Model (1) were replaced by the interaction effects of the factor with the factor trait. The latter resulted in a 2 × 2 variance covariance structure for each effect in Model (1) with a variance for each of the two response variables on the diagonal and a covariance on the off-diagonal.

Agronomic Efficiency (AE), an indicator of the increase in yield for units of fertilizer applied, was calculated as the slope of yield on applied fertilizer amount (Dobermann 2005). Therefore, treatment effects in Model (1) were replaced by a treatment-specific slope. To avoid over-parametrization, the slope for the control was set to zero. Thus, the intercepts serve as an estimate for the control, and slopes estimate the AE for all non-control treatments. The model was as follows:

$${y}_{ijkl}=\mu +{\beta }_{i}{x}_{ijkl}+{\varphi }_{j}+{\beta }_{ij}{x}_{ijkl}+{f}_{k}+{b}_{kl}+{e}_{ijkl},$$
(2)

where \({\beta }_{i}\) and \({\beta }_{ij}\) are the AE of treatment i and treatment i at crop j, and \({x}_{ijkl}\) is the fertilizer amount applied on plot l of field k crop j and treatment i. Afterwards, ratios were calculated and interpreted as an N fertilizer replacement value (NFRV) (Jensen 2013) (Eq. 3). The NFRV describes how efficiently a potential substitute for mineral fertilizers (here CAN) can provide available N for plants.

$$NFRV=100\cdot \frac{{\beta }_{i}}{{\beta }_{CAN}}$$
(3)

Models were implemented in the PROC MIXED procedure of the SAS version 9.4 software.

Results

Measurements during growth in 2020 (chlorophyll content assessment and biomass collection)

The 2020 SPAD values of the winter wheat were lower than the control treatment during the initial SPAD measurements in the LAS treatments, but exceeded the control over time (Fig. 1). Slurry, fermentation product, and mineral-N treatments were similar and showed the highest SPAD values. The values for winter barley showed similar effects of fertilizers in chlorophyll content (Table S2).

Fig. 1
figure 1

SPAD values of winter wheat (y-axis) at time in 2020 (metric x-axis) of measurements (22.04., 22.05., 05.06., 26.06.2020). Control: unfertilized treatment; CAN: calcium ammonium nitrate; LAS: liquid ammonium sulfate. Mean values of the five fertilizer treatments for each measurement date. For simplified presentation, the mean values were interpolated and are shown without comparison of means (given in Table S2)

Full size image

The treatments showed highly significant differences in the number and weight of spikes, weight of stems, harvested straw, and grain yield in the biomass cuts before harvest. The number and weight of spikes correlated positively with grain yield in both crops. In the biomass cuts, winter barley did not differ significantly from the control in stems and spikes. In winter wheat, by contrast, all treatments differed significantly from the control (Fig. 2A), where CAN, digestates, and manure had significantly higher yields than LAS.

Fig. 2
figure 2

A Dry matter means of biomass cuts of winter barley and winter wheat at BBCH 85. Winter barley was cut after 245 days of growth. Winter wheat was cut at 267 (F3) and 252 (F1) days of growth. B Mean straw and grain yields (t/ha) of winter barley and winter wheat in 2020. C Fresh matter (FM) and dry matter (DM) yield of maize and rapeseed grain in 2021. Note different scale for maize and rapeseed. Control: unfertilized treatment; CAN: calcium ammonium nitrate; LAS: ammonium sulfate (liquid). Within a crop and trait, mean values of fertilizer with at least one identical letter are not significantly different from each other (LSD, p < 0.005)

Full size image

Harvest in 2020 and 2021

In winter barley, the grain and straw yields of the CAN, digestate, and manure treatments were significantly higher than the LAS treatment and the control, with the grain yield of the LAS treatment significantly higher than the control. In winter wheat, LAS did not differ significantly from the control in both grain and straw yield, while the CAN, digestate and manure treatments had significantly higher straw and grain yields than both LAS and the control (Fig. 2B).

The factor crop type was significant for rapeseed and maize yields in 2021. The interactions of the factors treatment and crop were significant. In winter rapeseed, all fertilizer treatments had significantly higher yields than the unfertilized control. Whereas LAS and digestates did not differ significantly, CAN and manure had significantly higher yields than LAS. In maize, all treatments had significantly higher yields than the control but did not differ significantly from each other (Fig. 2C).

Nitrogen fertilizer replacement value (NFRV)

The NFRV of the fertilizers showed a high value for manure and digestate in all crops (Fig. 3; Table S4). LAS had the lowest values at 30% in winter barley and 40% in winter wheat. In the overall beneficial growing conditions of 2021, the NFRV of digestates and manure was higher than 100% in maize and rapeseed. LAS had an NFRV of 94% in maize and 75% in rapeseed.

Fig. 3
figure 3

Nitrogen fertilizer replacement values (NFRV) of ASL, digestate and manure for all crops (winter barley, winter wheat, maize, rapeseed) in both experimental years (2020/21). LAS: liquid ammonium sulfate. The NFRV was calculated as the ratio of Agronomic Efficiency. CAN (calcium ammonium nitrate) was used as the 100% mineral reference. The error bars show the standard deviation for each fertilizer per crop and year

Full size image

Discussion

How can scientific methods be successfully applied to a fertilizer experiment in an on-farm setting and scale?

In our participatory workshops with the farmers, it became clear that a randomized block design with all repetitions in one field ‒ typical for on-station experiments ‒ impedes the practicability of the experiment under on-farm conditions. For the farmers, and also the later interpretation of the results, it is important that the field management follows the farmers’ common practices as much as possible. It would be unfavorable, for example, if the farmers’ own manure or field equipment cannot be used, or if their tractors need to be turned frequently in the fields, as is the case for common block designs with many repetitions in one field. The latter may cause soil compaction with negative effects for the farmers even years after the end of the experiment. Where a single-field evaluation is not the aim, as in this study, Moehring et al. (2014) state that an alternative to treatment repetitions in one field is to have many fields acting as repetitions. In this case, it is better to have more fields and fewer replicates per field. In our experiment, the plots were designed in such a way that the farmers could drive from the beginning to the end of the fields using their conventional working width (Table 1; Figure S1). Because of the size of the plots, it was not possible to measure yield in the same way as in on-station experiments with small plot combines; instead, the methods described in Materials and Methods were applied. Here, it was advantageous that the farmers worked with their own machines under normal conditions, from which we expect more accurate data for agricultural practice. To validate our methods of measuring grain and straw yield, we compared the 2020 results with those of the manually harvested 2-m2 biomass cuts (weight and number of spikes, stem weight). The high correlations between the manually harvested biomass yields and those obtained by combines indicate that the results are comparable (Table S3). Making use of the local farmers’ knowledge in a participatory approach and the flexibility of the experimental set-up (Lightfoot and Barker 1988) were two major factors in attaining our on-farm experimental objectives.

A further important factor in the successful field trial performance was the fact that the scientists were present in person every time work or measurements were carried out in the fields. This enabled us to prevent potential mistakes and to intervene in a corrective way if necessary or possible. For example, at the very first fertilization, the randomization in the fields had to be renewed, because one fertilizer was applied incorrectly. Another advantage is that outliers in the data can be understood and explained. For example, we observed a late frost that was not recorded by our weather station [but confirmed by Strotmann (2020), LFL (2020)]. This damaged the winter barley fruit stalks, leading to a higher stem-to-tiller ratio visible in biomass cuts compared to winter wheat (Fig. 2A) and is thus a factor for lower grain yields (Fig. 2B).

The fact that we were able to establish cooperation at eye level, with an "all in one boat" mindset, was probably decisive for the smooth communication between the project partners. Lightfoot and Barker (1988) noted already 25 years ago that it is common for farmers to perform fieldwork that has not been agreed upon and that this can lead to data loss. Based on our experience, we assume that one reason for this could be that work is not always performed by the farmers themselves, but by family members or employees. This is not due to a lack of interest in the trial on the part of the farmers, but more to a lack of time. It is easier, for example, for an employee who is already on the road in the tractor with fertilizer spreader to quickly fertilize the trial plots. Here, it proved useful for us to maintain social contact with all people involved in the project to facilitate communication. In our opinion, the main reason farmers act spontaneously is that they adapt their work to weather conditions. As such, it is important that the project manager has practical experience and contacts the farmer in advance whenever the weather is optimal for procedures such as crop protection or soil cultivation, before he carries out any work in the field. In this way, it can be assured that the fields of all farmers are treated in the same way. The field management originally agreed upon may not be appropriate at the time of application due to sudden external impacts like weather, broken tools, etc.

It can be concluded that close contact and exchange with the farmers in a participatory approach is essential for the successful performance of an on-farm experiment with many farmers. Further, it is necessary for the project manager to have practical agricultural knowledge to understand the circumstances the farmers work in, and to be always present when fieldwork takes place. This constant presence strengthens the social fabric among the project partners and helps to explain possible outliers in the data. Instead of repetitions in single fields, an experimental design with many fields, each serving as a block, allows farmers to use their own farming techniques and is as precise as a randomized block design in the statistic evaluation. Because the experiment was designed for practical farming, the results can be transferred directly into the farmers' practices. In this study, this was even the case during the project period.

What is the most suitable application technology for LAS in farming practice?

Ammonia losses of LAS

Observations made during the SPAD measurements showed that the plants did not grow very much in height in 2020 due to the drought and so did not develop much straw. Later, however, after the onset of flowering, there was sufficient rainfall (Table S1). The yields (Fig. 2B) show straw-to-grain ratios far below the typical values of 0.33 to 0.61 (Dai et al. 2016). It is very likely that the water deficit between early plant development and flowering negatively influenced the growth of the stems and therefore the straw yield (Shao et al. 2008; González et al. 1999; Samarah 2005; Samarah et al. 2009). The lower yields (Fig. 2B) and SPAD means (Fig. 1) of the LAS treatments in 2020 compared to 2021 also indicate that the nitrogen applied in 2020 did not reach the plants. Compared to the other fertilizers where part of the nitrogen was bound in organic matter (manure fermentation products) or in the CAN granules, the ammonium sulfate solution was taken up directly by the soil. Jones et al. (2013) state that less ammonia dissolves in warmer than in cooler soil water. This could be a reason for higher LAS volatilization losses. However, since the soil was already in a poor condition due to the drought at the time of the first fertilization, we assume that water and with it the LAS ammonia evaporated directly. Soil temperature and soil pH are driving factors for ammonia losses from the soil solution. If the pH rises above 7, the relationship shifts from ammonium to ammonia (Jones et al. 2013). The pH values of the soils in this on-farm experiment were between 6 and 7.34 after harvest in 2020, with no differences between treatments. Wester-Larsen et al. (2022) investigated the volatilization potential of the LAS used in this experiment under controlled laboratory conditions. They state that, in sandy soil, LAS had lower volatilization than most of the other fertilizers (novel biobased fertilizers) they studied. However, they did not test it in silty clay loam soils, the soil type present in our experiment. The soil pH values between pH 6 and 7.34 in our experiment and the influence of the parent carbonate rock present may have been additional drivers of ammonia losses to the atmosphere. The pH of the ammonium sulfate solution used in our experiment fluctuated around 7.8 due to the production process, and therefore, as an alkaline fertilizer, could carry an increased risk of N losses by itself. We consider the high pH of the LAS used as the driving factor for volatilization losses during application. Compared with the other fertilizers, the LAS treatments had significantly lower yields in the drought year 2020 but similarly high yields in 2021 when growing conditions were generally good (Fig. 2). These contrasting results make it clear, that the environment has a major influence on the effectiveness of the fertilizer. In the meantime, we were able to change the LAS production process so that its pH can be lowered to 6.5. We expect this to reduce the risk of losses and thus increase the availability of nitrogen to the plants.

It can be concluded that, compared with other fertilizers like mineral fertilizer granules that only dissolve in contact with moist soil or fertilizers where part of the N is organically bound, the mineral fertilizer LAS seems to have a higher volatilization risk due to its liquid form and pH. Further studies are needed to establish the pH value of liquid ammonium sulfate that leads to the least volatilization losses in specific soil types.

Osmotic stress?

The LAS SPAD values of winter wheat (Fig. 1) show that wheat did not have more, and barley had less chlorophyll (Table S2) than the control treatment after the first fertilization. Manurecomine (2016) reported the death of their violets after the application of ammonium sulfate. This was due to osmotic stress, which caused the plants to take up less water and the roots to "chemically burn". Jones et al. (2013) and Sigurnjak et al. (2019) did not find this effect for ammonium sulfate in their review of other studies nor in their own experiments with lettuce and maize. However, the authors note that, in their own studies, there was adequate rainfall/irrigation after fertilization. The assumed high water and LAS ammonia evaporation caused by the complete lack of precipitation in our on-farm experiment, even weeks after the first N application in March, may have caused solutes (sulfate) to concentrate in the soil of the plots. In studies where air scrubber wastewater (such as LAS) was investigated, the authors note that the high electrical conductivity (EC) value of these recycled fertilizers, indicating an increased salt concentration, may potentially cause salt stress during fertilization (Vaneeckhaute et al. 2013a; Alburquerque et al. 2012; Sigurnjak et al. 2016). Much of the circumstantial evidence described above suggests that in our trials salinity in LAS treatments could have resulted in osmotic stress, as this is known to suppress plant growth (Gupta and Huang 2014). Symptoms of stress were visible in the thin, yellow plants on the fields and measured in lower SPAD means compared to the unfertilized control (Fig. 1; Table S2). However, our study was not technically designed to investigate salt stress, as this was not one of the objectives. Nevertheless, it should be taken into consideration that salt stress could occur in farm practice through LAS application if there is no precipitation after fertilization or if the LAS is not diluted (as was the case in 2021) but applied pure (with three-jet nozzles in 2020). So, although it was not clearly demonstrable, our results indicate ASL could cause osmotic stress to crops in practical arable farming when applied to very dry soils.

Most suitable application technology

In this on-farm experiment, the task was to integrate a novel fertilizer into the farmers' existing field management system, using their existing farm technology. The most convenient method for farmers to apply LAS in 2020 was using a field sprayer with three-jet nozzles. Through the LAS jet, the fertilizer was expected to pass the leaves, reach the soil, and then penetrate the soil to be dissolved in the soil water. Due to suspected salt stress and the significantly lower yields of the LAS treatments in wheat and barley in 2020, it was clear in 2021 that the application technique should be improved. In a long-term study on fertilization application techniques, which also took place on soils in Hohenlohe, surface LAS application (drag hose) was compared to injection (star wheel) into the soil (Mokry 2013). The author states that injecting LAS into the soil as a fertilizer depot is the best option to reduce losses such as ammonia volatilization. Unfortunately, the necessary star-wheel technology was not available in 2021. However, the farmers had a slurry tanker with a dragging hose/trailing shoe available for applying the liquid manure and digestates. Through the trailing shoe applicator, LAS should be scratched into the soil to minimize contact with the atmosphere. However, the conveyor technology of the slurry tanker is too voluminous for precise fertilization with pure LAS. Therefore, LAS was diluted with water to adjust the delivery rate of the system to the N concentration. In this way, the appropriate amounts corresponding to plant requirements could be fertilized in a targeted manner. The maize fields were fertilized before sowing, but rapeseed fields were fertilized during crop growth (Figure S1d). Here, no relevant plant losses were determined. Based on experience from the two years of the experiment, we were able to derive a fertilization technique suited to the farmers’ real working conditions.

As a fertilizer relatively sensitive to environmental impacts, LAS should be applied in a manner that reduces these impacts (e.g. volatilization losses). For the farmers participating in this project, the “slurry tanker-trailing shoe applicator” solution was practicable. A possible future scenario could be that farmers purchase the star-wheel applications service from a contractor or the LAS manufacturer itself. Therefore, further studies are required to make an economic comparison of the slurry tanker with dragging hose and star-wheel technology.

Can the recycling fertilizer LAS serve as a substitute for CAN in common fertilizing regimes?

Nitrogen fertilizer replacement value

In a review of field assessments, Schils et al. (2020) compared 10 studies that calculated the nitrogen fertilizer replacement value (NFRV) of organic fertilizers based on yield and N uptake of different crops and application techniques. Six of these used CAN as a reference. None of these studies resulted in a higher NFRV than 83%. These findings stand in contrast to our results (Fig. 3) where most values were higher, the highest being 157% for digestates in maize. One explanation for this is that in our on-farm experiment we did not compare the effect of individual fertilizers on yield, but the effect of fertilizing systems as defined in the fertilizer legislation (DüV 2020) (Table 2). The organic fertilizers were complemented by CAN as additional mineral compensation. In 2020, the digestate treatments received ~ 1/3 to 2/3 of total N in the form of CAN and the manure treatments received ~ 1/2 of total N in the form of CAN. In 2021, the organic treatments did not receive any CAN in maize, while in rapeseed 1/3 to 1/2 of total N was applied through CAN.

One main reason for the high NFRV in our on-farm experiment with manure and digestate (> 125%) in 2021 could be that the NFRV in the second year is an accumulated value. By calculating the AE, we calculate a ratio between references. In our study, the reference’control’ did not receive any fertilizer in either years. Thus, in the second year, the total N in the control soil is probably even lower than in the first year. This stands in contrast to the organic fertilizers ‘manure’ and ‘digestates’ which were applied in both years. Through the application of organic fertilizers, the N pool in the soil is enriched over the years, while there is nearly no residual response with mineral fertilizers (Riley 2016) like CAN and LAS. When fertilizing according to the German fertilizer legislation (DüV 2020), 10% of the N that was fertilized with organic fertilizers in the previous year must be taken into account (deducted) for fertilization in the following year. However, this is only a rule of thumb. Organic nitrogen from previous organic fertilization may have accumulated and been plant available above the 10% of the previous year's fertilization. To precisely determine the accumulated organic N Pool in the soil, a second control (not fertilized in 2020, but fertilized in 2021) could have been established. This idea was excluded for reasons of practicability in the on-farm experiment. In future projects, it would be worthwhile investigating the accumulated value of fertilizers, if the limitations of the experimental design allow. Furthermore, the LAS application technology was changed from three-jet nozzles in 2020 to a slurry tanker with a trailing shoe applicator (rapeseed) and pre-seed incorporation in maize. The change to an application technique where the fertilizer is dissolved in water and placed/scraped directly onto/into the soil may have led to fewer volatilization losses compared to the three-jet nozzle technique in 2020.

It can be concluded that the fertilization systems of liquid manure and fermentation products with compensatory fertilizers according to the German fertilization ordinance can be very good possibilities to replace systems based on pure mineral fertilization in practical agriculture. However, in order to compare the NFRV of fertilizer systems as they are applied in practice, long-term trials are needed that take into account the accumulated organic N (and P) pool in the soil.

LAS as a substitute for CAN

Overall, the NFRV results show that the combination of organic and mineral fertilizers can lead to similar or better results than with mineral fertilizer alone. The results of 2020 and 2021 (Figs. 2, 3) show that LAS + TSP does not produce higher yields than organic fertilizers + CAN or its mineral counterpart CAN + TSP. In drought conditions, LAS—as a liquid fertilizer—presumably loses N to the atmosphere faster than the CAN granules, which in 2020 only dissolved when the weather became more humid.

Based on our experience with LAS, we suggest using this fertilizer as a mineral compensatory fertilizer on top of organic fertilizers in accordance with the German fertilizer legislation (DüV 2020). LAS can potentially substitute CAN, but as a single fertilizer on its own, it offers less yield security than CAN. In a field experiment with maize, where LAS was used as a balancing/compensatory fertilizer, LAS achieved a 5% higher N fertilizer replacement value (NFRV) in a pig manure/LAS/potassium sulfate treatment compared to a pig manure/CAN/potassium sulfate treatment (Sigurnjak et al. 2019). The fertilizers were incorporated into the soil by hand. In agricultural practice the problem is that, as a liquid fertilizer, LAS cannot be applied as easily as is the case with granules e.g. CAN. While granules can be applied during the whole growing season using a tractor with special tires and a spreader, the slurry tanker with a dragging hose or even star-wheel technology used for LAS limits applicability during the growing period, depending on the crop. If a crop is already well developed, it may be damaged beyond a threshold acceptable to farmers. This restricts the application window to the early growing stages of a crop. However, two application timeframes are still possible. To achieve high yields or qualities, we suggest considering nitrification inhibitors so that the N of the second fertilization becomes plant-available at a later growing stage after flowering to fill the grains of the crop.

In conclusion, as an environmentally sensitive fertilizer, LAS cannot serve as a substitute for CAN on its own, but as a mineral balancing/compensatory fertilizer on top of organic fertilizers. The farmers in this study were able to integrate LAS into their fertilization practice as a substitute for CAN. In order to make precise statements about the fertilization effect on different environments in practical farming, further on-farm studies would be necessary.

Conclusion

The outcomes of the regular meetings with the farmers (e.g. that plot length should be linear to the field length) and close social contact proved essential for the success of this on-farm experiment. Due to its liquid form and the fact that all N is present as ammonia, liquid ammonium sulphate proved to be more sensitive to environmental impacts that can limit the fertilizing effect than mineral fertilizer granules (calcium ammonium nitrate) or semi-liquid fertilizers where part of the N is organically bound (manure, digestate). Therefore, liquid ammonium sulphate should be applied in a manner that reduces these impacts such as injection or incorporation. For the farmers participating in this project, the “slurry tanker-trailing shoe applicator” solution was practicable. Overall, the NFRV shows that the combination of organic fertilizers and mineral fertilizers can achieve similar or higher results than fertilizing systems that rely on mineral fertilizers only. As an environmentally sensitive fertilizer, liquid ammonium sulphate cannot serve as a substitute for calcium ammonium sulphate on its own in practical arable farming but as a mineral balancing/compensatory fertilizer on top of organic fertilizers.