Struvite forced precipitation has gained attraction since the 90’s, not only to avoid infrastructure damage but also as a P recovery technique . This process has been studied and improved in the past years making it a more efficient precipitation process . Although the production of struvite is gaining popularity, its commercial production is still scarce. The potential of P delivery of a WWTP in the form of struvite in the system where this study is located has been previously quantified by Rufí- Salís et al. , demonstrating the potential of these widespread installations to provide this ill distributed resource. In terms of application, the properties of struvite as an effective source of nutrients  for plants and its low solubility in water  make it a slow-releasing valuable fertilizer that can reduce economic costs in agriculture . However, only limited literature has explored the application of struvite in agricultural facilities. For example, Antonini et al. , Uysal et al. , Gell et al.  and Liu et al. assessed the maize performance of struvite with different characteristics and origins in different soils. In a review made by Li et al. we can see that almost all struvite trials found that vegetables grown with struvite had the same -or even improved- performance compared to controls with conventional fertilizers. Creating a closed-loop, waste-to-resource system such as that of struvite recovery within the city limits and not applying it at this scale seems contradictory within the concept of urban metabolism. In this sense, the synergy between struvite precipitation in urban WWTPs and urban agriculture seems worth exploring considering the potential of the latter to blur the lines between waste and resource within urban areas . This article aims to assess the potential of struvite precipitated in a WWTP as a fertilizer within the framework of urban metabolism.

Based on experimental and analytical results performed on a Phaseolus vulgaris crop grown in a hydroponic rooftop greenhouse, we determine the implications of fertilization with struvite in terms of yield, vertical farming racks water flows and P balances and provide recommendations to further improve the performance of this waste-to-resource fertilizer.Struvite granules were obtained from Aarhusvand A/S company from Aarhus, Denmark. This company distributes fertilizer grade struvite under the name PhosphorCare™, recovered using the Phosphogreen™ technology . This technology is based on a fluidized bed reactor that creates the specific conditions to precipitate struvite through the addition of magnesium chloride, sodium hydroxide and air. The final struvite granules have a size range of 0.5–1.5 mm. Common bean plant was chosen as the crop for this study, planting nursery plants . To apply the struvite to the plants, we considered different possibilities. Mixing it with the nutrient solution was discarded because the system could not benefit from the slow-release characteristics of struvite. Thus, we choose to directly apply the granules to the plant roots. Considering this option, we designed a system that consisted on mixing perlite with struvite inside a low-density polyethylene perforated bag with holes of no more than 1 mm diameter . At the same time, this system allows the interaction between struvite granules and roots and avoids the loss of undissolved struvite into the leachates due to draining through the perlite bag. Two different experiments were carried out: the validation test and the determination test, both of them using double growing lines with 8 substrate bags each . The validation test served as a previous experimental set-up to determine if the proposed methodology was functional and correct possible influencing variables in the experiment, such as the use of the plastic bag to retain the struvite close to the plant rhizosphere or to scale the most suitable quantities of struvite for the determination test. On the other hand the determination test was designed with the previous experience of the validation test.All other mineral fertilizers were maintained.

The initial nutrient concentration of the substrate was verified to be negligible at the beginning of the experiment through atomic spectroscopy and elemental analysis. Samples of the fertilizer solution were collected directly from the drippers placed in the perlite bags.To determine the PNS and PLIX, the respective samples were collected three times per week and externally analysed using ICP-OES atomic spectroscopy . PSI was quantified summing the amount of perlite in a specific bag with the amount of struvite that was applied, considering weights obtained by drying two struvite samples and two perlite samples at 105 °C in a furnace until reaching constant weight . PSF was quantified differently in each test. In the validation test, all 4 samples for a specific treatment were homogenized after extracting the roots, using distilled water to separate the struvite granules from the roots. After this process, two random samples were dried at 105 °C in a furnace until reaching constant weight, then grounded and digested with concentrated HNO3 in a Single Reaction Chamber microwave and externally analysed using ICP-OES atomic spectroscopy. On the other hand, in the determination test, roots were shredded, homogenized and integrated within every individual substrate sample. Then, a fraction of these samples was dried and analysed using the same method as in the validation test. PLV, and PST were determined based on the nutrient content of every plant separately. Leaves and stem were separated, sorted into paper envelopes and dried in a furnace at 65 °C until reaching constant weight , grounded and digested with concentrated HNO3 in a Single Reaction Chamber microwave before analyzing externally the concentration of P through ICP-OES atomic spectroscopy. The same methodology was applied to determine the PBN, with randomly chosen 500-gram bean samples being processed for every treatment. The measured P content of beans was multiplied by the measured rates of biomass production to estimate the rate of P accumulation in crop biomass.From September 13th to December 3rd, 2018, 10 double growing lines were used , distributing the treatments as showed in Fig. SM2 of the Supplementary material.

The aim of this experiment was to validate and keep track of different parameters of the system, like for instance, make sure that the small, perforated bag did not have negative consequences on the crop development. To do so, we split the control lines into two different treatments, VCB and VC0, using standard nutrient solution with and without the bags, respectively. Secondly, to check the correct development of bean plants with struvite in a hydroponic system, we applied different struvite amounts per plant: 5, 10, 15, 20 and 25 g corresponding to the treatments tagged as V5, V10, V15, V20 and V25, respectively. Additionally, a treatment with no struvite was tagged as V0. These amounts of struvite were based on previous experiments done with the same crop species and variety in hydroponic cultivation that accounted for P uptake . One week after the first harvest, KPO4H2 was added in the nutrient solution of struvite treatments until the end of the harvest to ensure a good nutrition to the plants during the production period, which is highly demanding in P .The production results for the control treatments VCB and VC0 showed that the perforated bag did not have any effect on the correct crop development and yield , as the yields from the different lines do not differ between them . Even though treatment VC0_1 generated more yield , it could be attributed to the fact that it was an exterior cropping line facing the border and thus received more radiation. Similarly, VCB_2 also produced more yield than its replicate although no significant differences where determined by the end of the experiment. On the other hand, treatments with struvite exerted a similar yield than the control treatments at the end of the crop. The treatment with the highest quantity of struvite had the highest production median , while the treatment with the lowest quantity of struvite had the highest mean . On the other hand, the treatment without struvite produced a really low yield . The similarities in terms of yield between all struvite treatments at the end of the cycle may be related to the addition of KPO4H2 fertilizer during the production phase. Moreover, we can see that struvite treatments produced more than the control in the first 3 harvests. This effect is similarly observed for the phenological stages . For the parameters that were quantified in different dates , side shoots , open flowers and floral buttons, we can see that the treatments with struvite not only had a correct early stage development, but also develop plant organs earlier than in control treatments.

The V0 treatment doesn’t show P results in leaves for 54 and 78 DAP because no leaves remained in the plant at the sampling time. For this same reason, there is a lack of data in beans for 78 DAP. Finally, concentration in beans for struvite treatments was similar to the one observed in the control. For all plant organs, a pattern in the accumulation of P in the plant tissue can be observed. In the first sampling all treatments show a rather low accumulation with greater content for plants with greater struvite quantities, in some cases also for the control treatments. For the second sampling, a bigger content difference can be seen with an acute increase of the V25 P content, especially for the stems and leaves. Finally, at 78 DAT, these differences between treatments even out and only treatment V0 remains significantly reduced. This last part however, does not correspond to the undissolved P in the perlite, where the P content in the substrate directly responds to the amount of struvite given, being always higher for the V25. The control treatments receive the P through irrigation making the existing content in the substrate comparably small.Figs. 3  shows the results of the accumulated yield per number of harvests, being the sixth harvest the final one before uprooting the plants. Only treatments S1  and S2.5 had lower yields than the control treatment , the first being significantly lower. On the other hand, all other treatments with 5 g of struvite or above produced more than the control treatment, demonstrating the potential of struvite to produce similar or even higher yields than with mineral fertilizer, as reported by Li et al. . As we can see in Fig. 3, it was not until the second harvest  that great differences were observed between the S1 yield and the other struvite treatments, while a decrease in S2.5 yield was observed between the 4th and 5th harvest, 57 and 64 DAP, respectively. Regarding the control treatment, vertical rack system the first harvest produced lower yield  than the S5 struvite treatment being even similar to the treatments with the lowest struvite application S1 .

This fact reinforces the idea that the application of struvite could be benefificial for early stage plant development, as the validation test showed better behaviour in struvite than in control in phenological variables. This fact could be related to the NH4+ supply by struvite, which could benefit the plant root balance when combined with nitrate supply . The fact that previous literature suggests that NH4+ supply to common bean could be harmful for plant development could be related to the amount of NH4+ supplied. Because struvite does not only enable a slow release of P but also of NH4+, reaching NH4+ accumulation to harmful levels seems improbable. In terms of distribution, yields show an asymptote behaviour among treatments, where S20 produces the highest yield and S1, the lowest . Fig. SM15 in the Supplementary material shows how treatment S10 was detected as the exception for this tendency in terms of mean production,probably related to bias parameters like shapes in the greenhouse or a nonhomogenic distribution of struvite in the perlite bag. However, boxplots represented in Fig. SM16 shows how the median of the final amount of yield harvested for S10  follows the tendency, while not presenting outliers in the distribution.Fig. SM17 of the Supplementary materials shows that the irrigated water in the control and the struvite treatments was the same , while Fig. 4 shows the accumulated P during the entire cycle in the different water streams.