Field experiments were carried out at two organically managed AFS in Switzerland where both climates are classified as Cfb (Köppen and Geiger classification), i.e. warm and temperate climate. The first AFS (10.2 ha) is located in Windlach in the Northern part of Kanton Zürich (47°32′16.4′′N, 8°28′46.3′′E), 410 m a.s.l. Apple trees (Malus domestica, cv. ‘Heimenhofer’, ‘Schneiderapfel’ and ‘Spartan’) were planted in a density of 37 trees ha−1 (10 × 28 m distance within and among tree rows, respectively) in West–East orientation in November 2015. Windlach has a mean annual temperature of 9.2 °C and a mean annual precipitation of 1038 mm with the summer months (June to August) receiving the highest amounts (111–123 mm per month) (https://de.climate-data.org/europa/schweiz-221 2018). The second AFS (1.9 ha) is located in Seegräben in the central part of Kanton Zürich (47°20′4 7.1′′N, 8°45′42.0′′E), 536 m a.s.l. Cherry trees (Prunus avium, cv. ‘Avione’, Prunus cerasus, cv. ‘Königin Hortense’ and ‘Hallauer Aemli’) were planted in a density of 62 trees ha−1 (8 × 20 m distance within and among tree rows, respectively) North–South orientation in 2011. Seegräben has a mean annual temperature of 8.7 °C and a mean annual precipitation of 1130 mm with the summer months (June to August) receiving the highest amounts (128–135 mm per month) (https://de.climate-data.org/europa/schweiz-221 2018). Both farms differ in their management: While tillage is reduced in Seegräben, it is intensive in Windlach due to weed control. Grown crops are diverse vegetables, summer crops (Panicum sp., Triticum aestivum subsp. spelta) and winter wheat in Seegräben and sunflower (Helianthus annuus), squash (Cucurbita sp.) and winter rye (Secale cereale) in Windlach. In Windlach, artificial pasture and fallow occupy several years in the crop rotation. Details can be found in the Supplementary Information.
On July 22, four random soil samples were taken with a cylindrical soil tube at each site and further divided in samples of 0–10 cm and 10–20 cm depth. Soil carbon and nitrogen were measured with the CHN628 Series Elemental Determinator at ETH Zurich. For total phosphorus determination, wet digestion with H2SO4 after Anderson and Ingram (1993) was used and samples where then analysed on the ICP. All soil nutrients differed significantly between sites. Soil carbon content was significantly higher in Seegräben (average 2.92% ± 0.28 SD) than in Windlach (average 1.59% ± 0.28). Likewise soil nitrogen content was significantly higher in Seegräben (average 0.29% ± 0.03) than in Windlach (average 0.14% ± 0.03), as was phosphorus (average 890 ± 111 mg P kg−1 in Seegräben and 636 ± 150 mg P kg−1 in Windlach). Soil nutrient levels in Seegräben correspond well to the average soil C, N and P levels present in agricultural soils in Switzerland (C: 3.13%, N: 0.29%, P: 932 mg kg−1, source: NABO, pers. communication), but are rather low in Windlach.
In spring 2019, two shade intensities were implemented by means of artificial shade nets (R.G. Vertrieb, Austria) to achieve moderate and heavy shading. Shade net intensities were 40% (moderate) and 90% (heavy) of full sunlight (i.e. 60% and 10% of incident photosynthetically active radiation). However, the area beneath the shade nets received lateral sunrays during early and late hours of the day (as they would in a natural stand of trees). These degrees of opacity were chosen based on previous studies, where yield reduction was observed by a 30–50% relative irradiance (Dufour et al. 2013; Dupraz et al. 2018a) and model results suggest a 55–75% (17 m wide tree alleys) and 25–50% (35 m wide tree alleys) light capture by trees (Dupraz et al. 2018a). For a steeper contrast, severe shading was selected for the heavy shade treatment. Control treatments had no shade net. Nets measured 2 × 2 m and their height was adjusted throughout the season (0.3–1.2 m) to shade the central plot (0.7 × 0.7 m). Barley was sown on 2 April in Seegräben and 28 March in Windlach.
Shade (40%, 90%) and control (0% shade) treatments were examined in a full-factorial design with four treatments: (1) irrigation, (2) fertilisation, (3) fertilisation and irrigation and (4) control. The four environmental treatments with the three shade intensities built one replicate (12 plots per replicate). In Windlach there were four replicates (48 plots in total), in Seegräben there were three replicates due to limited space (36 plots in total).
Irrigation was conducted manually with watering cans. The amount of water was adapted to weather conditions and varied between 9 and 27 l per plot (i.e. 4.5–18.5 mm m−2) once a week. Throughout the barley growing season in 2019 approximately 81 l per plot were irrigated in Seegräben and 90 l per plot in Windlach.
For fertilisation, the organic mineral NPK-fertiliser “Styria Fert Veggie Plus P + S” (Agro Power Düngemittel GmbH) was used in Windlach. It is manufactured out of residues from starch and glucose production, cocoa pods, Hyperphosphate, elemental potassium and sulphur and consists of 4% N, 5% P2O5 P and 2% K2O with 7% CaO, 0.45% Mg, 5% S, 0.527% Fe and 0.0218% Zn. Styria Fert Veggie Plus has a C:N ratio of 10:1 and a pH of 6.5. Its dry matter has 65% organic substances. It is certified after the BIO AUSTRIA guideline and the Council Regulation (EC) No 834/2007 for organic agriculture. 520 kg ha−1 were applied on 22 April 2019. In Seegräben one half of the land used for the experiment received high nutrient input in the form of compost the autumn before. Due to a mistake during the sowing of the understorey crop, the area with the fertilisation treatment was initially sown with spelt and then ploughed again to remove the spelt before sowing of barley. However, the ploughing could not completely eliminate the germinating spelt. Therefore, the factor fertilisation is confounded with the unwanted spelt growth. This was accounted for during data analysis (see “Data analyses” section).
The gross plot area was 2 × 2 m, however only the central area of 0.7 × 0.7 m (0.49 m2) was manually harvested when plants reached maturity stage (end of July in Windlach, beginning of August in Seegräben 2019). Where the central plot area was damaged by accumulated water at heavy rainfall events on the surface of the shade nets and subsequent centralised trickling or by lodging, another area within the gross plot was sampled. The sampling position was documented as either “central” or “border” and taken into account during data analyses. The harvested material (ears with short stalks) were put in labelled paper bags and stored in a dry room for 21–28 days at the ETH Research Station for Plant Sciences in Eschikon (Lindau) where they were threshed with the threshing machine “Saatmeister Allesdrescher K35” (rotational frequency: tumbler: 9, fan: 6). Grains were weighed and stored in small paper bags in a dry room. For trait measurements, four individuals within the area designated for harvest were randomly selected at harvest time. The whole aboveground plant was manually harvested and put in labelled bags made out of baking paper which were left open to ensure drying. Internode length between nodes were measured separately and noted down and the number of stalks counted (tillering). Plant height, a common indicator for growth under shade, was obtained by adding all internode lengths, measured from four individuals per plot at harvest. Total grain yield was weighed, the total number of seeds counted and the seed mass calculated by randomly weighing 10 seeds and dividing the weight by 10. Straw (air-dried aboveground stalks) was weighed at the end.
Statistical analyses were carried out with R version 3.6.1. The data was tested for normality and homogeneity of variance by the Shapiro–Wilk test, the Fligner–Killeen test and a visual inspection of residuals. On the plot level, a logarithmic transformation was carried out for yield in the general model and the Windlach-model. On the individual level, a logarithmic transformation was applied for total seed number. In the plot-data one outlier (strong lodging occurrence) and in the individual-data one outlier (extremely high yield) was removed. Differences in group means among groups was analysed by multifactorial ANOVA (type I, sequential sum of squares). Significances of each factor were assessed by means of the F-test. Statistical modelling was performed with three linear models on the plot level – one model for each experimental site and a general model for both locations. The formula of the Seegräben-model is lm(yield ~ spelt + sampling position * plant density * shade * irrigation) where “yield” combines barley and spelt yield, “spelt” is the presence (1) or absence (0) of spelt in the experimental plots, “sampling position” is the sampling position (central/border) of the harvested area (70 × 70 cm) within the shaded area (2 × 2 m), “plant density” is the number of plants within the harvested area, “shade” is the applied shade treatment (0, 40%, 90% shade) and “irrigation” refers to the received irrigation treatment (yes/no). The factor “fertilisation” (yes/no) is not included in the Seegräben-model as it is inseparable with the presence of spelt. Similarly, the Windlach-model is lm(log(barley yield) ~ sampling position * plant density * shade * fertilisation * irrigation) with barley yield being log transformed. The general model at plot level was lm(log(yield) ~ site + sampling position * plant density * shade * fertilisation2 * irrigation) with “site” being the experimental site (Seegräben/Windlach). Again, the response variable “yield” includes barley and spelt yield and is log-transformed. As the factors fertilisation and presence of spelt are confounded in Seegräben, a second fertilisation factor (“fertilisation2”) was created with “yes”, “no” and “X” where “X” compiles those plots in Seegräben which were fertilised and where spelt was present. Models on the individual level followed the same structure, accounting additionally for dependency of individual samples within the same plots in linear mixed effects models of the structure lme(plant trait ~ spelt + site + sampling position + plant density + shade * fertilisation * irrigation, random = ~ 1|plot). Models on the individual level (Windlach data only) followed the same structure, accounting additionally for dependency of individual samples within the same plots in linear mixed effects models: lme(plant trait ~ sampling position + plant density + shade * fertilisation * irrigation, random = ~ 1|plot). Multiple coefficients of determination for the general model and the Seegräben- and Windlach-model were 0.91, 0.82 and 0.89, respectively, and lower for the models on seed mass, total seed number and height on the individual level (0.41, 0.57 and 0.46). For post-hoc analysis a Tukey test was used to compare the means of treatment groups with the HSD.test()-function within the R agricolae package (de Mendiburu 2020) with a significance threshold of α = 0.05. Partial effects of each factor were extracted by means of the effect()-function within the R effects package (Fox and Weisberg 2019) for analysing main and interaction effects.