Introduction

Maintaining functional and biodiverse urban green spaces is fundamental for liveable cities (cf. SDG 111). Urban gardens are a major component of urban green spaces in many countries2,3. They are heterogeneous in structure, but despite their relatively small size they provide critical habitat resources and increase the connectivity of urban landscapes4. Garden management creates diverse garden land-use types including perennially vegetated habitats such as lawns or annually vegetated habitats such as vegetable beds2. These diverse microhabitats support urban biodiversity and have the ability to provide nature’s contributions to people5,6. The worldwide increase in human population is expected to take place mainly in urban areas7, while growing cities often expand onto fertile agricultural soils, thus challenging the supply of fresh food in the future8. There is a great potential for producing food in urban gardens and at the same time to provide other ecosystem services (ES) in densely populated cities9. It is estimated that urban farming delivers food for approximately 800 million people10, although the current global scale is difficult to assess11. However, hundreds of millions of citizens rely on urban agriculture for part of their nourishment12. Nonetheless, urban garden soils are also important for regulating soil functions such as water storage (flood control13), C and N storage14, pollination15, soil formation16, pest control17, or to decrease urban heat island intensity18 and provide habitats for many species even in densely urbanised areas6. From a sociological perspective, urban gardens are important for recreation, well-being, and social interaction19.

Urban gardening has a long tradition in many countries around the world20. As a consequence of decades of beneficial soil management practices, such as the application of compost21, urban garden soils may not always be as poor in quality and potentially polluted as other urban soils3,22. Despite the importance of gardens for urban biodiversity6, information on the ecological importance of allotment and domestic gardens is still scarce compared to public green spaces2,23. However, there is a large body of evidence that biodiversity drives ecosystem processes and related services in aboveground communities24, but the functioning of belowground biodiversity is much less understood25. Although it has been shown that soil biodiversity is linked in multiple ways with aboveground biodiversity26,27, further investigation is needed to better understand these relationships. Garden soils are strongly influenced by human activities3,22,28, but they are also affected by the past land-use, the degree of disturbance or climate related drivers such as the urban heat island effect29. Soil functions are provided and controlled by a large variety of soil organisms30, also in urban soils28, where the frequency of soil disturbance is often high29. Changes in community composition of soil fauna in both alpha and beta diversity31, for instance due to soil disturbance, can impair soil functions such as organic matter decomposition or nutrient retention32. The interactions of aboveground and belowground species, driving ecosystem functions, at least at the local scale33, are mainly linked via plants34. However, still very little is known about this relationship between aboveground and belowground diversity and associated soil functions35, especially for garden soils36.

The ability of an ecosystem to provide multiple functions, so-called multifunctionality24, can be calculated as indices based on the functions of interest37. Such measures of multifunctionality (i.e. the averaging approach), have been used to analyse a wide range of ecosystem drivers38, such as soil characteristics39, habitat diversity40, climate41, or management practices in agriculture42 and even in constructed ecosystems such as green roofs43. Here, we focus on five independent measurements for calculating soil multifunctionality ranging from aboveground44 and belowground27 litter decomposition, to nutrient supply for plant growth45 and water regulation, such as water storage capability46.

Research on urban garden soils has recently received increased attention3,17,28,47,48, especially with regard to human health and well-being29. However, our understanding of the complex interactions between management practices, soil biodiversity and soil functioning is still scarce49. In this study, we focus on gardening activities in the two most dominant garden types of Zurich (CH), allotment and domestic gardens, and assess the interactions between aboveground diversity of plants and belowground diversity of soil fauna. We investigated earthworms (Oligochaeta: Lumbricidae), representing soil macrofauna species and springtails (Hexapoda: Collembola), representing soil mesofauna species, as indicators for soil functioning50 and assessed the impacts of urban gardening on soil multifunctionality. Earthworms are generally described as ecosystem engineers50, due to their impact on soil structure and quality, at least in temperate soils51. They are important indicator organisms for soil functions52, soil disturbance, and management practices50. It has been shown that also in urban ecosystems such as parks or urban gardens, they are sensitive indicators of anthropogenic management intensity47. Springtails are a key group of microarthropods50 and can be used as indicators of sustainable land-use, soil quality27, or the use of pesticides53. Moreover, they are used to assess soil functionality54 and the impact of environmental factors55 on soil biodiversity. In addition, we assessed soil microfauna by biological soil measurements, such as basal respiration, microbial biomass and gene copy numbers of bacteria and fungi (Table 1).

Table 1 Soil characteristics describing the soil quality of urban garden sites used as explanatory variables in the SEM.

The overall objective of our study was to investigate impacts of garden management practices (management intensity index, garden land-use types) on aboveground plant diversity and belowground diversity of soil fauna, and their direct and indirect effects on soil multifunctionality. We hypothesised (cf. a priori structural equation model (SEM) Fig. 1) that (i) intensive soil management will reduce the diversity of both plant and soil fauna and negatively affect soil multifunctionality (arrows 1 & 2). We assumed that (ii) aboveground and belowground diversity are also linked in urban garden ecosystems and therefore expected that a higher diversity of plants will have a positive effect on both soil fauna and soil multifunctionality (arrows 3). Furthermore, we expected that (iii) soil fauna diversity (arrow 4a) and biomass (arrow 4b) will have a direct positive effect on soil multifunctionality. Additionally, we assumed an influence of (iv) soil characteristics and (v) urbanisation on soil multifunctionality (arrows 5 & 6).

Figure 1
figure 1

A priori SEM model with hypothesised direct and indirect effects of urban gardening on soil multifunctionality. Expected positive relationships are given in black and negative ones in red, grey arrows represent both positive and negative effects. We expected that annual vegetables (arrows 1a) will negatively influence plant and soil fauna as well as soil multifunctionality compared to perennial grass sites, while perennial flowers (arrows 1b) will show positive effects. Management intensity (arrows 2) is expected to negatively affect plant diversity and soil fauna as well as soil multifunctionality. Higher plant diversity (arrows 3) is hypothesised to have a positive effect on soil fauna and soil multifunctionality. Soil fauna diversity and biomass (arrows 4a & 4b) are also expected to have a positive effect on soil multifunctionality. Urbanisation (arrows 5) might have a positive or negative effect on soil fauna and soil multifunctionality. Expected effects of soil characteristics (arrows 6) can be found in Fig. S9.

In a second step, we analysed soil fauna community structure. We expected that frequently disturbed soils would have the lowest species diversity within (alpha diversity) and among (beta diversity) garden sites, including a high community evenness and beta diversity mainly driven by species loss (nestedness) rather than species replacement (turnover). For the plant community, we expected highest alpha and beta diversity for garden sites with high planting activities, including a high species turnover component for beta diversity. Furthermore, we investigated impacts of management practices on soil fauna community composition and on soil fauna disturbance indices.

Results

Urban gardening effects on soil fauna and soil multifunctionality

The SEM based on our a priori expectations (Fig. 1) of urban gardening effects on aboveground and belowground diversity and soil multifunctionality met the criteria of Fisher’s C statistic56 (Fisher’s C = 30.7; p = 0.80; AICc = 286.8). The model included one significant missing path57 between PC1 and soil fauna diversity (SEM; 0.18; p = 0.03). With the inclusion of this path the overall model fit of the SEM improved (Fisher’s C = 24.3; p = 0.93; AICc = 288.3), with marginal differences in the AICc (1.5). Overall, the strongest relationships in the SEM originated from garden land-use types, influencing plant diversity, fauna biomass and soil characteristics (PC2) and soil multifunctionality both in indirect and direct ways (Figs 2 and S10). The strongest effects on soil multifunctionality came from soil PC1 (SEM; −0.61; p =< 0.001), represented by lower loads of Cmic, Corg, bacteria, Fe and K, but higher soil bulk density values (Fig. S4). Soils with increased Cmic, Corg, bacteria, Fe and K, but lower bulk density values thus covaried with higher soil multifunctionality. Annual vegetable sites showed lower soil multifunctionality values (SEM; −0.40; p = 0.03) compared to perennial grass sites. Moreover, we found positive effects of plant diversity (SEM; 0.17; p = 0.01), and fauna biomass (SEM; 0.17; p = 0.02) on soil multifunctionality. Taken together, both significant and non-significant effects explained 74% of the total variation of soil multifunctionality. In addition, we also identified several indirect effects on soil multifunctionality (Fig. 2 and S10, Table S6). We found that plant diversity had a positive indirect effect on soil multifunctionality mediated by increased fauna diversity and fauna biomass. Plant diversity itself was positively affected by flower & berry sites (SEM; 0.37; p = 0.04) and negatively by management intensity (SEM; −0.22; p = 0.01), explaining 39% of the variation in plant diversity. A similar pattern was found in high beta diversity values (Table S5) for the plants (0.94 ± 0.001), dominated by a high turnover component (0.92 ± 0.001) and low nestedness component (0.02 ± 0.001), indicating the high variability between garden plots. Moreover, the management intensity indirectly negatively affected soil multifunctionality by decreasing fauna biomass and plant diversity (Fig. 2, Table S6). Plant and soil fauna beta diversity and fauna phylogenetic diversity were not included in the final SEM (cf. Table S10, Fig. S7) due to a large increase in the AICc (588.5) and because it explained only 3% more variance in soil multifunctionality (Fig. S8).

Figure 2
figure 2

Final most parsimonious SEM connecting garden management (land-use types, management intensity), urbanisation, plant and soil fauna diversity with soil multifunctionality (AICc = 288.3, Fisher’s C = 24.3, P = 0.93). Arrows represent unidirectional relationships among variables. Black arrows denote significantly (p < 0.05) positive and red arrows significantly negative relationships (Table 3). Dashed grey arrows represent non significant relationships (p > 0.05). The thickness of paths has been scaled based on the magnitude of the standardised regression coefficient. Conditional R2s, based on the variance of both the fixed and random effects, as well as marginal R2s, based on the fixed effect parts for each component models are given in the boxes of the response variables. Soil multifunctionality consists of five measurements related to important soil functions. Soil characteristics are included in Fig. S10.

Urban gardening effects on soil fauna community composition

We investigated the effects of management practices, plant ecological indicators (Table 2), soil characteristics (Table 1), garden land-use types, and urbanisation on the community composition of both earthworms and springtails (Table 4). Both soil fauna groups were strongly affected by the plant ecological indicator nutrients (PERMANOVA EW; F = 5.9; p =< 0.001, COL; F = 5.4; p =< 0.001), that represented the requirements of nutrient-rich soils for plants that are predominantly found in vegetable sites (Fig. 3). Earthworm communities were further affected by the penetration resistance of the soil (PERMANOVA EW; F = 5.3; p =< 0.001), favouring anecic species such as L. terrestris. Furthermore, plants indicating humus rich soils (PERMANOVA EW; F = 3.0; p = 0.01) favoured the two most abundant earthworm species (Table S1, Fig. 3) A. chlorotica and A. caliginosa. Those soils were further associated with higher contents of Mg (PERMANOVA EW; F = 2.1; p = 0.04) and K (PERMANOVA EW; F = 2.7; p = 0.01), resulting from higher soil disturbance (PERMANOVA EW; F = 2.4; p = 0.02) predominantly in vegetable sites. In summary, the NMDS ordination was driven by endogeic species A. chlorotica (NMDS; R2 = 0.30; p =< 0.001) and A. caliginosa (NMDS; R2 = 0.13; p =< 0.001), anecic species L. terrestris (NMDS; R2 = 0.17; p =< 0.001), and both endogeic (NMDS; R2 = 0.20; p =< 0.001) and anecic (NMDS; R2 = 0.19; p =< 0.001) juveniles. Although the garden land-use type was not a significant factor affecting the earthworm community composition, the most abundant species can be allocated to specific land-use types (Fig. 3, Table S1). In vegetable sites we primarily found endogeic species such as A. chlorotica (64.5%), which is tolerant to disturbances28 and a pioneer species58, endogeic juveniles (41.9%) and A. caliginosa (45.1%). Perennial grass sites were dominated by anecic species such as L. terrestris (52.5%) and anecic juveniles (47.6%), probably due to deeper and more compacted soils. As expected, we found the lowest earthworm diversity (DSimpson) in vegetable sites (1.85 ± 0.2) and the highest in grass sites (2.57 ± 0.2). The earthworm diversity was driven by endogeic and anecic species but not by epigeic species (Table 4). Additionally, we found the lowest beta diversity (βJAC) for earthworms in vegetables and the highest in grass sites (Fig. S5, Table S5). However, evenness (ESimpson) was not highest in vegetable sites but in grass sites and flower & berry sites, where we also observed higher nestedness components (βJNE).

Table 2 Garden management practices based on the gardener survey (Table S3) and plant ecological indicator values reflecting the plant environmental requirements81.
Table 3 Final most parsimonious structural equation model (SEM; AICc = 156.3, Fisher’s C = 24.3, P = 0.93) indicating direct and indirect effects on soil multifunctionality from garden land-use types, garden management, plant and soil fauna diversity, soil fauna biomass, soil characteristics and urbanisation.
Table 4 PERMANOVA of earthworms (PERMANOVA EW; left) and springtails (PERMANOVA COL; right) and management practices, plant ecological indicators, soil characteristics and garden characteristics as explanatory variables.
Figure 3
figure 3

Soil fauna community structure analysis of earthworms (a,b) and springtails (c,d). Soil fauna species are coloured in grey, corresponding to three ecological categories (Table S1). Only species with a significant (p < 0.001) effect (Table S4) and only significant explanatory variables (p < 0.05) from the PERMANOVA model (Table 1) were fitted. Garden land-use types include standard deviation bars.

Springtails were affected by the plant ecological indicators moisture (PERMANOVA COL; F = 3.7; p =< 0.001) and moisture variability (PERMANOVA COL; F = 3.7; p =< 0.001). These indicate moist soil or alternating soil moisture conditions, favouring species that were more abundant in grass sites such as S. aureus (81.1%) or P. alba (67.3%) on soils with high penetration resistance (PERMANOVA COL; F = 2.3; p = 0.01) and Cmic (PERMANOVA COL; F = 2.1; p = 0.02). All three life forms of springtails were present in the species that had the biggest effect on the community composition (Table S4). B. hortensis showed the largest effect (NMDS; R2 = 0.29; p =< 0.001), and was most often found in vegetable sites (75.5%), where also C. thermophilus (NMDS; R2 = 0.13; p =< 0.001) was often present (56.7%), correlating with potassium loads (PERMANOVA COL; F = 2.5; p =< 0.001). Other species driving the community composition of springtails were mostly found in grass sites, such as P. notabilis (85.4%), representing the most abundant (22.9%) springtail species (13,435 individuals) in this survey. Moreover, we found eight springtails (marked with stars in Table S1) which were not included yet in the Fauna Europaea species list, with two new records for Switzerland (I. balteatus and I. graminis) according to the available literature and expert opinion (c.f. Table S1). The separation of flower & berry sites in the NMDS was mainly driven by C. denticulata (NMDS; R2 = 0.08; p =< 0.001). Moreover, we found a clear effect of garden land-use type (PERMANOVA COL; F = 2.8; p =< 0.001), but also two significant effects of specific garden management practices: applying water (PERMANOVA COL; F = 2.5; p =< 0.001) and weeding (PERMANOVA COL; F = 1.8; p = 0.04). Weeding was more attributed to vegetable sites and applying water to grass sites in the NMDS ordination, whereas flower & berry sites were associated with a higher degree of urbanisation (PERMANOVA COL; F = 1.8; p = 0.04) and with more alkaline soils (basification; PERMANOVA COL; F = 2.0; p = 0.02). As expected, we found lower mean values for the springtail diversity (DSimpson, Table S2) in vegetable sites (3.3 ± 0.2) compared to grass sites (3.8 ± 0.1). Beta diversity (βJAC) was highest for flower & berry sites with a high turnover in comparison to the nestedness component (Fig. S5, Table S5). Springtail evenness (ESimpson) was highest in flower & berry sites, where we also found the highest nestedness component (βJNE), probably due to hemiedaphic and euedaphic species being more similar in flower & berry sites.

Additionally, we found differences between garden land-use types in soil fauna disturbance indices. The collembolan ecomorphological index and the earthworm anecic to endogeic ratio were lowest in vegetable sites (Table S2). The acari to collembola ratio was lowest in grass sites and the fungal to bacterial ratio was highest in vegetable sites.

Discussion

Worldwide, there is a growing interest of city administrations in the socio-economic and ecological benefits of urban gardens6,11,12,17,23,29,59. We investigated impacts of garden management practices on aboveground and belowground diversity and interlinked soil functions. The SEM (Fig. 2, Table 3) revealed direct effects on soil multifunctionality and indirect effects mediated by soil fauna. Overall, our results showed that the largest effects on soil multifunctionality were caused by specific soil characteristics. Soils showing high biological soil quality indices such as organic and microbial carbon and bacteria increased the potential for soil multifunctionality. This probably originates from organic gardening practices such as the application of compost, due to the correlation with increased potassium loads and with decreased bulk density values (Table 3, Table 1), also influencing soil quality22. The second strongest effect on soil multifunctionality was caused by the cultivation of vegetables and legumes in annual vegetable sites (hypothesis (i), Fig. 1 arrows 1a), probably due to the frequent soil disturbance and the unprotected open soils in comparison to perennial grass sites. The cultivation of flowers and berries increased plant diversity (Fig. 1 arrows 1b), but decreased soil fauna biomass compared to grass sites. Urban gardens with higher plant diversity (hypothesis (ii), Fig. 1 arrows 3) increased soil multifunctionality directly, and indirectly through increasing fauna diversity and thus fauna biomass. The general pattern of enhanced soil multifunctionality with increased plant diversity is in line with results found in other ecosystems such as croplands, shrublands, grasslands, and forests, where plant diversity increased ES such as pollination, C storage, pest control, and productivity60,61. Contrary to our expectations, we found no significant direct effect of management intensity on soil multifunctionality, but more intensively managed sites decreased plant diversity and fauna biomass. A similar relationship of management intensity and decreased diversity has been observed in urban lawns. Lerman et al.62 showed that mowing only every three weeks instead of every week increased the numbers of flowers by 2.5 times and thus the abundance and diversity of bee populations. Although Tresch et al. showed that aboveground63 and belowground22 organic matter decomposition increased with urbanisation, there was no significant effect of urbanisation (hypothesis (v), Fig. 1 arrow 5), on soil multifunctionality.

The structure of earthworm and springtail communities were influenced by plant ecological indicators (Fig. 3, Table 3), representing the living conditions of plants. Interestingly, springtails were more affected by plant ecological indicators than earthworms, highlighting the dominant influence of plants on springtails50. As expected, we found a lower alpha diversity of earthworms and springtails in vegetable sites, likely due to the high soil disturbance. Beta diversity was constantly high in both soil fauna and plant communities, driven by high turnover and low nestedness components. The plant community composition was shaped by the high species turnover between the garden sites, with highest dissimilarities for flower & berry sites. As expected, these significant differences originated most likely from planting and other garden management practices leading to site specific community compositions. The beta diversity values for both fauna communities were lowest in vegetable sites and peaked for earthworms in grass sites and for springtails in flower & berry sites, reflecting the different ecological strategies of earthworms and springtails.

Earthworms are important indicators for soil functioning51,52. Functional groups of earthworms have been used to detect impacts of cultivation in different soils such as pastures, orchards or forest soils, while the ratio of anecic to endogeic species was used as an indicator of contaminated soils64, or soil disturbance65. In the frequently disturbed vegetable sites we found the lowest values for the earthworm anecic to endogeic ratio and the collembolan ecomorphological index, indicating a decreased soil biological quality due to soil disturbance48. Other studies have reported the highest value of the collembolan ecomorphological index in urban vegetable gardens and forest sites48. However, the highest value found in forest sites (2.3) was still considerably lower than the average we found in urban gardens of Zurich (5.8 ± 0.1), with a high number of euedaphic springtails such as P. pulvinata, F. quadriculata, and I. minor. Springtail abundance often increases from agricultural to forest sites66. Here they increased form vegetable to grass sites by a factor of 4.3. Besides soil disturbance, the increased abundance in grass sites could be explained by the higher plant cover of the perennially vegetated sites66 and because grass strips offer a variety of microhabitats for soil mesofauna species48. In contrast, Joimel et al.48 found higher mean densities in vegetable beds than forest or grassland sites, underlining the quality of urban gardens for soil fauna biodiversity and soil quality. In addition, the increased organic matter content in urban garden soils3 can be an important factor for the high soil fauna diversity, since the input of organic matter in garden soils can be higher than in agricultural fields67. Moreover, high management intensity is known to decrease soil mesofauna diversity68. We found a lower acari to collembola ratio in grass compared to flower & berry sites, which is in line with the dominance of acari in frequently disturbed arable or vineyard soils48. Additionally, we found a higher springtail biomass for grass sites, while earthworm biomass was at a similar and comparably high level in all urban garden land-use types. For instance, earthworm abundance (227.4 ± 15.5) was considerably higher than mean reference values for biological soil quality indicators found in grass or cropland soils69. Referring to all microorganisms68, Cmic peaked for grass sites, while the mean value (780.9 ± 21.3) was higher than in cropland soils (341 mg kg−1), but lower than in grassland soils (1249 mg kg−1) found in Belgium69 or Switzerland70 (2077 mg kg−1). This pattern of Cmic reflected management practices such as fertilisation or tillage68. Additionally, the composition of soil microorganism communities is an important driver for soil functioning35. For instance, a shift in fungal composition or activity can increase carbon uptake and nutrient cycling35. Both soil disturbance71 or lower plant diversity72 can result in decreased fungal to bacterial ratios. While increasing fungal to bacterial ratios can be expected from desert to temperate grassland and forest soils, assuming that grassland soils are more bacteria dominated than forest soils26. Here, we found an increased fungal to bacterial ratio for vegetable sites, due to the increased fungal and decreased bacterial gene copy numbers in those sites (cf. Table S9, Fig. S13). This might be related to the input of compost on the vegetable sites or the increased plant diversity compared to grass sites.

The intuitive and rather simple concept of multifunctionality24 and its reduction to one single metric, such as the averaging approach37, needs to be examined critically. For example, the functions and methods to measure them must be carefully selected42. The strength of the biodiversity ecosystem multifunctionality relationships depends on the number of included functions, which was generally stronger when more functions were considered73. Another point is that the aggregation of multiple functions into one single metric can obscure information about potentially contrasting single functions74. The highest correlations among the soil functions (cf. Fig. S14) were found between Cmin and Nmin (r = 0.45, p =< 0.001), both used to calculate soil nutrient supply, and between Cmin and WHC (r = 0.43, p =< 0.001). All other correlations (r < 0.27) claimed a certain independence of the selected soil functions. The moderately positive correlation of all components to soil multifunctionality is required, because negative correlations among functions can be a limitation for multifunctionality assessments73. However, this multifunctionality assessment framework could also be used in future studies to assess the impact of managed urban green spaces on nature’s contributions to people in cities.

With this city-wide assessment of the effect of urban gardening practices on aboveground and belowground diversity of plants and soil fauna, we demonstrated the potential impacts of gardeners’ decisions on the quality and functioning of the soil and implications on the biodiversity of a city. In conclusion, our study suggests that a higher plant diversity can directly or indirectly increase soil multifunctionality by enhancing soil fauna diversity and biomass. In a previous study, intensive garden management decreased soil quality indices22. Here we demonstrated that a high garden management intensity indeed also declined plant diversity and soil fauna biomass, with negative impacts on soil multifunctionality. In addition, we analysed drivers shaping soil fauna community structure of earthworm and springtail species. We showed that both were affected by plant ecological indicators, soil characteristics, and management practices such as the frequency of soil disturbance or applying water. We conclude that increasing plant diversity together with soil protective management practices have the potential to increase soil functions as well as foster biodiversity, and to create more biophilic59 urban gardens, supporting human well-being and the ecological value of urban green spaces. Even though soil is a key resource in cities, it has not been integrated in most urban green space plans75, thus we recommend that urban gardens including ecological management practices should be integrated in future green city strategies.

Methods

Study design and gradients

This study took place in 85 urban gardens of the city of Zurich, Switzerland (Fig. S1). We selected gardens based on three independent criteria22,76: (i) the type of garden (domestic N = 43 vs. allotment; N = 42 Fig. S2), (ii) the management intensity (such as intensively managed vegetable or flower beds or extensively managed meadows), and (iii) the degree of urbanisation, ranging from densely built-up to peripheral areas within the city boundaries. In each garden two sampling plots (2 m × 2 m) with different land-use management were selected (Table S12), belonging to one of the following three categories: annual vegetable beds (vegetables; N = 47), perennial flowers and berries (flowers & berries; N = 52) or perennial lawn and meadows (grass; N = 71), reflecting the most dominant garden land-use types in Zurich and in many other cities.

Garden management practices were assessed using a questionnaire with 26 management questions, specific for each land-use type, ranging from the frequency of lawn cutting to fertiliser application or weeding (Table S3). Garden management intensity was assessed as the sum of 26 management questions. In addition, five common management practices (disturbance, fertiliser, pesticides, water, weeding; Table 2) were used in the community composition analysis. Urban warming was used as a proxy for urbanisation due to the correlation with the amount of built-up and paved area for different radii (30–500 m) around the gardens63. It has been assessed as the deviation in local mean air temperatures at night near the surface based on a local climate model77, showing temperatures increased of up to 5 °C for urbanised gardens.

Aboveground diversity

Plant diversity was assessed by a floristic inventory78 of cultivated and spontaneously growing plants on each sampling plot (N = 170). Based on this inventory of 600 plant species, we calculated plant alpha diversity as the total number of plant species per sampling plot and plant beta diversity as the mean of the pairwise Jaccard dissimilarity comparisons between each focal plot and all other plots79. Additionally, we used a six-point ordinal scale78 to calculate community weighted mean values of seven plant ecological indicator values80 (Table 2), reflecting the plant environmental requirements81.

Belowground diversity

Earthworms were collected in a smaller subplot of 0.3 m × 0.3 m within the 2 m × 2 m sampling plots by a combined hand sorting and mustard extraction method22. Earthworms were stored in 70% ethanol58, identified to the species level, and classified into three ecological categories (Table S1): epigeic species (living in the litter layer, with little burrowing activity), endogeic species (living in the soil, with horizontal burrows) and anecic species (living in large and deep vertical burrows).

Springtails and mites were sampled with six undisturbed soil cores (5 cm diameter, 8 cm length, Eijkelkamp, NL) randomly taken in the 2 m × 2 m sampling plots63. Springtails were identified to the species level including life forms according to ecological and functional traits (Table S1): epedaphic species (living in the upper litter layer), hemiedaphic species (living at the interface between litter and soil) and euedaphic species (soil-dwelling species).

We defined soil fauna diversity as the average proportional species richness across soil macrofauna (earthworms) and mesofauna (springtails) species following Allan et al.82. Soil fauna beta diversity was calculated as the average proportional species beta diversity of earthworms and springtails, while the individual measures of beta diversity per soil fauna group were computed as mean pairwise Jaccard dissimilarities, similarly to the plant beta diversity. Soil fauna biomass was calculated as the average proportion of biomass per m2 of soil, with measured earthworm biomass [gm−2] on an individual basis (including gut contents) and estimated springtail biomass (conversion factor of 5 g for each springtail83).

Soil fauna disturbance indices

The adaptation of soil fauna to management practices was assessed with four soil fauna disturbance indices: the collembolan ecomorphological index48, the acari to collembola ratio84, the fungal to bacterial ratio26, and the earthworm anecic to endogeic ratio64.

Soil characteristics

Soil characteristics were assessed with a combination of three physical, six chemical and four biological soil measurements (Table 1), representing the most commonly used soil quality indicator measurements85. The microbial community information of bacterial (16S) and fungal (18S) gene copy numbers were used to calculate the fungal to bacterial ratio. Measurement details can be found in Table S13 and Tresch et al.86.

Soil multifunctionality

Similar to other studies87 we used the averaging approach37 to calculate soil multifunctionality. It calculates the mean value across standardised soil functions for each sampling plot. In total, we used five measurements (Table S11), which are related to important soil functions, for the computation of soil multifunctionality. The three assessed key soil functions are (i) aboveground and belowground litter decomposition, (ii) soil nutrient supply, and (iii) soil water storage and regulation. The soil function litter decomposition aboveground was measured by standardised leaf litter mass loss (Zea mays L.) in 4 mm mesh sized litter bags63, while belowground litter decomposition was measured by the net mass loss of green tea bags, buried in 8 cm soil depth22. The supply of nutrients in the soil was assessed by the mineralisation rates of N (Nmin) and C (Cmin), and the capacity of the soil for water regulation, was measured by the water holding capacity (WHC).

Data analysis

Soil fauna diversity and biomass were calculated by taking species richness per taxonomic group, applying a standardisation for each taxonomic group scaled to a range from 0 to 1 (\(f(x)=({x}_{i}-{x}_{{\min }})\)/\(({x}_{{\max }}-{x}_{{\min }})\)) and then averaging the values for each plot82. Aboveground and belowground beta diversity were calculated as mean pairwise Jaccard dissimilarities comparing each focal plot to all other sampling plots79 using the R package ‘betapart’88. Soil multifunctionality was computed by scaling each of the five measurements of soil functions to a range from 0 to 187 and deriving mean values across the standardised soil functions according to the averaging approach37. Community weighted means of plant ecological indicators were calculated with the R package ‘FD’89.

We fitted a piecewise structural equation model (SEM), with the ‘piecewiseSEM’ package57, to infer relative importance of direct and indirect effects of urban gardening, plant diversity, urbanisation and soil characteristics on soil fauna and soil multifunctionality. To address multicollinearity and reduce the amount of variables we applied a PCA for the soil characteristics and used the first four PCA axes, explaining 64.2% (Table 1; Fig. S4) of the variation (Kaiser-Guttman criteria). We used Shipley’s d-separation test to identify missing paths in the SEM and the AICC for model comparison. We used linear mixed effect models (LMEM; lme(nlme;90) with the garden as random effect for each SEM component and reported standardised (scaled by mean and variance) path coefficients, as well as marginal R2 and conditional R2 based on fixed and random effects57 (Table 3). Model assumptions were tested (Fig. S11) and potential spatial autocorrelation patterns were calculated with Moran’s I autocorrelation indices and the spatial structure in the model residuals using semivariograms (Fig. S12).

We applied individual LMEM with garden identity as random effect and land-use types as response variables to assess changes in fauna and plant diversity and soil fauna disturbance indices. We checked for normal distribution, autocorrelation, and heteroscedasticity of the model residuals and applied a transformation (\(\mathrm{log}(x+1)\)) in the cases of: earthworm biomass, anecic to endogeic ratio, acari to collembola ratio and springtail biomass. We reported means and 95% credible intervals of the Bayesian inference posterior distribution based on 10,000 independent simulations91. Soil fauna community structure was further analysed using a permutational multivariate analysis of variance (PERMANOVA, 10,000 permutations) with a Hellinger transformed Euclidean distance species matrix of earthworms (EW) and springtails (COL) and a non-metric multidimensional scaling (NMDS) using the ‘vegan’ package92. For the NMDS only significant variables from the PERMANOVA were fitted. Data management and statistical analyses are provided as an R project using R 3.4.2 (R Core Team, 2017).