Impaginato 123 Adv. Hort. Sci., 2023 37(1): 123­132 DOI: 10.36253/ahsc­13899 Sustaining low­impact practices in hor­ ticulture through non­destructive approach to provide more information on fresh produce history and quality: the SUS&LOW project M. Amodio 1, G. Attolico 2, L. Bonelli 3, M. Cefola 4, H. Fazayeli 1, F.F. Montesano 3, 5, B. Pace 4, M. Palumbo 1, 4, F. Serio 3, A. Stasi 1, G. Colelli 1 (*) 1 Department of Science of Agriculture, Food and Environment, University of Foggia, Via Napoli, 25, 71122 Foggia, Italy. 2 Institute on Intelligent Industrial Systems and Technologies for Advanced Manufacturing, National Research Council (CNR), Via G. Amendola, 122/O, 70126 Bari, Italy. 3 Institute of Sciences of Food Production, National Research Council (CNR), Via Giovanni Amendola, 122/O, 70125 Bari, Italy. 4 Institute of Sciences of Food Production, National Research Council of Italy (CNR), c/o CS‐DAT, Via Michele Protano, 71121 Foggia, Italy. 5 Department of Soil, Plant and Food Sciences, University of Bari, Via Aldo Moro, 70126 Bari, Italy. Key words: Marketing strategies sustainability, non­destructive assessment, quality, shelf­life. Abstract: The general aim of the project SUS&LOW is to increase the sustainabil­ ity of fresh produce by testing and implementing low­input agricultural practices (LIP) with positive impact on product quality with the support of non­destructive (ND) tools for real­time quality assessment and for product discrimination. Additionally, new marketing strategies are generated to better support the added value of the products and to satisfy the final consumers’ preferences. The SUS&LOW project consists of three work packages (WP) and the adopted methodology used two model crops: rocket salad and tomato. The WP1, focused on the reduction of agricultural inputs, showed that sensor­based fertigation management might improve sustainability of soilless cultivation. Results coming from WP2, aimed to the evaluation of ND techniques, outlined the high poten­ tiality of hyperspectral imaging (HSI) and Fourier transformed­near infrared (FT­ NIR) techniques for the authentication of sustainable growing methods. Moreover, project activities’ proved computer vision system (CVS) as an effec­ tive tool for evaluating the product quality also through the bag. The WP3, deal­ ing with marketing strategies, indicated a positive approach of consumers com­ pared to LIP products certified through a visual storytelling platform. 1. Introduction Production of vegetable crops under controlled environments (i.e. greenhouses) has expanded considerably over recent decades in (*) Corresponding author: giancarlo.colelli@unifg.it Citation: AMODIO M., ATTOLICO G., BONELLI L., CEFOLA M., FAZAYELI H., MONTESANO F.F., PACE B., PALUMBO M., SERIO F., STASI A., COLELLI G., 2023 ­ Sustaining low‐impact practices in horticul‐ ture through non‐destructive approach to provide more information on fresh produce history and quality: the SUS&LOW project. ­ Adv. Hort. Sci., 37(1): 123­132. Copyright: © 2023 Amodio M., Attolico G., Bonelli L., Cefola M., Fazayeli H., Montesano F.F., Pace B., Palumbo M., Serio F., Stasi A., Colelli G. This is an open access, peer reviewed article published by Firenze University Press (http://www.fupress.net/index.php/ahs/) and distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Competing Interests: The authors declare no competing interests. Received for publication 26 October 2022 Accepted for publication 10 January 2023 AHS Advances in Horticultural Science https://doi.org/10.36253/ahsc-13899 http://www.fupress.net/index.php/ahs/ http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ Adv. Hort. Sci., 2023 37(1): 123­132 124 Mediterranean areas (FAO, 2013). Initially, research efforts and the related introduction of technical inno­ vations focused on high­quality, healthy products. However, concern with environmentally­sustainable production has risen in the last decade as industrial greenhouse crops are usually seen as entailing high environmental impact (Torrellas et al., 2012). On the other hand, there is also plenty of evidence that greenhouse vegetable production may decrease the environmental impact compared to the field cultiva­ tion (Stanghellini, 2014). The efficient use of resources (water and fertiliz­ ers), in irrigated greenhouse agriculture, is a promis­ ing and increasingly adopted strategy to achieve bet­ ter crop performance, improved nutritional and sen­ sorial quality (Montesano et al., 2015; Montesano et al., 2018). With respect to traditional systems, soil­ less cultivation and, particularly, closed­cycle with recycling of nutrient solution (NS) produce a number of benefits, including the possibility to standardize the production process, to improve plant growth and yield, and to obtain higher efficiency in water and nutrients use. In addition, it is also possible to modu­ late the regulation of the secondary metabolism of plants through an optimal control of the nutrient solution composition, or by imposing controlled stresses, or through biofortification treatments, gen­ erally leading to an improvement in the nutritional value of products (Rouphael and Kyriacon, 2018; Renna et al., 2022). Innovative technologies based on the use of sensor networks for fertigation manage­ ment may considerably reduce water and fertilizers consumption and increase the overall use efficiency of those inputs, and may lead to qualitative and quantitative improvements while preventing both under­ and over­irrigation. The most used instrumental techniques to mea­ sure quality attributes of fruits and vegetables are destructive and involve a considerable amount of manual work, primarily due to sample preparation. In addition, most of these analytical techniques are time consuming and sometimes may require sophisti­ cated equipment. Finally, they can be performed only on a limited number of specimens (samples) and therefore their statistical relevance may be limited (Amodio et al., 2017 a). Research has been focused on developing non­contact, rapid, environmental­ friendly, and accurate methods for non­invasive eval­ uation of quality in fruits and vegetables. Nowadays, there are a few emerging non­destructive analytical instruments and approaches for this task, including spectroscopy, hyperspectral imaging, and computer vision (Liu et al., 2017). Near infrared spectroscopy has gained wide atten­ tion in the food sector due to its capacity of providing fingerprints of different products on the base of the interaction between their molecular structure and the incident light (Workman and Shenk, 2004) which is the result of different pre­harvest factors that also affect the final composition and quality. The feasibili­ t y o f N I R S ­ b a s e d a n a l y s i s t o e v a l u a t e q u a l i t y attributes of fresh fruits for commercial application have been reported by numerous authors (Arendse et al., 2018; Palumbo et al., 2022 a). Hyperspectral imaging (HSI) combines the princi­ ples of spectroscopy and conventional imaging or computer vision. It is mainly used for internal bruise and defect detection in fruits and vegetables (Ariana and Lu, 2010; Babellahi et al., 2020; Tsouvaltzis et al., 2020) but also to predict the internal composition (Piazzolla et al., 2013; Yang et al., 2015; Liu et al., 2017; Piazzolla et al., 2017). Amodio et al. (2017 a) showed the potentiality of hyperspectral imaging in the Vis­NIR spectral range to predict internal content of soluble solids, phenols, and antioxidant activity of fennel heads. In addition, this technique provided important information about the maturity of fennel heads which may be used to determine the optimal harvest time. Some studies successfully applied these methods for the discrimination of production origin and agricultural practices, as revised in Amodio et al. (2020). NIR and HIS were in fact used for the classifi­ cation of apples (Guo et al., 2013), persimmon (Khanmohammadi et al., 2014), and arabica coffee (Bona et al., 2017) from different origins. As for pro­ duction systems (Sánchez et al., 2013) investigated the potentiality of NIRS technologies to discriminate green asparagus grown under organic and conven­ tional methods. More recently, Amodio et al. (2017 b) successfully discriminated conventionally and organically grown strawberries, being also able to identify two different types of organic production systems applied to the same genetic material on the same site, soil, unheated tunnel. All these studies have suggested multispectral and hyperspectral systems as valid tools to evaluate qual­ ity of different agricultural products and, more inter­ estingly, as tools for product authentication. Finally, Computer Vision Systems (CVS) may be applied to extend quality prediction and discrimina­ tion along the whole supply chain from harvesting up to consumers. CVS combine mechanics, optical Amodio et al. ‐ The SUS&LOW project 125 instrumentation, electromagnetic sensing, and digital image processing technology (Patel et al., 2012). Recently, CVSs have been used to assess quality and marketability of tomatoes (Arias et al., 2000), arti­ chokes (Amodio et al., 2011), fresh­cut nectarines (Pace et al., 2011), fresh­cut lettuce (Pace et al., 2014), fresh­cut radicchio (Pace et al., 2015), and rocket leaves (Cavallo et al., 2017). Moreover, they have been applied for the prediction of internal qual­ ity of colored carrots (Pace et al., 2013). Even more interesting is the application of these systems during the post­packaging phase and along the whole distri­ bution chain. Despite the relevance of quality evalua­ tion of packaged products, few investigations were reported in literature. Multi­spectral reflective image analysis has been applied to monitor the evolution and spoilage of leafy spinach covered by plastic materials (Lara et al., 2013); more recently, Cavallo et al. (2018) have proposed an application of image analysis by CVS for non­destructive and contactless evaluation of quality of packaged fresh­cut lettuce. Therefore, the interest of investigating the applica­ tion of CVS to detect quality and shelf­life of pack­ aged products. Finally, the possibility of using non­destructive technique for increasing the information on product history (e.g. growing location and agricultural prac­ tices) may be considered as baseline to develop mar­ keting tools to promote the diffusion of sustainable production system. Cost barrier is an obstacle for choosing low input products instead of the conven­ tional, even if environment is mentioned as a strong commitment (Krystallis and Chryssohoidis, 2005). Therefore, the knowledge about consumer prefer­ ences for the adoption of LIP is still matter of debate. The general aim of the project is to increase the amount of sustainably­produced food by testing and implementing low­input agricultural practices with positive impact on product quality with the support of non­destructive tools for real­time quality assess­ ment and product discrimination, which may inspire new marketing strategies to better support the added value of the products and increase incomes of potential users. 2. Project activities and main results The SUS&LOW project structure consists of three work packages (WP). WP1 focused on research activi­ ties aimed to reduce agricultural inputs (water and fertilizers) in greenhouse cultivation, chosen as a strategic high­value sector for Mediterranean agricul­ ture. This WP was also in charge of making available to the project team vegetables products (rocket and tomato) different for the level of sustainability char­ acterizing the cropping system adopted, to be used in other WPs for the related investigations. Then, WP2 was aimed to the quality assessment and to the implementation of new tools to acquire information about quality and history of fresh produce obtained with LIP (WP1). Non­destructive methods (including NIR, hyperspectral imaging and image analysis by CVS) have been used for food authentication, show­ ing interesting and promising results. Finally, WP3 realized ad hoc survey to analyze the consumer behaviour with respect to the possibility of purchas­ ing fruit and vegetables LIP certified (WP1) and iden­ tified by ND technologies (WP2) with the aim to implement adequate marketing strategies. In this section, an overview of the research strategies and approaches adopted in the three WPs is provided. The main results are reported and discussed. WP1: quality crops through low‐impact practices Based on the overall project structure, this WP was focused on soilless cultivation, since it has the potentiality to achieve extremely high water and fer­ tilizers use efficiency, beside high yield and quality, in intensive cropping systems. However, the adoption of free­drain open cycle with empiric fertigation schedule management operated by timers (the pre­ dominant case in Mediterranean area), may compro­ mise the sustainability of soilless culture. Therefore, the adoption of strategies aimed to rational use of water and fertilizers and excess leaching prevention is a key­factor for increased sustainability and reduced environmental impact of soilless culture (Massa et al., 2020). In this context, substrate mois­ ture/EC (electrical conductivity) sensor­based irriga­ tion is a promising and increasingly adopted strategy to reduce water and fertilizers consumption and loss­ es, and to improve the overall crop performance, product quality and production process sustainability in soilless greenhouse cultivation (Palumbo et al., 2021 a). Several experiments were carried out at the Experimental Farm La Noria (Mola di Bari, BA) of the CNR­ISPA (Bari), with the common approach to com­ pare treatments providing traditionally adopted empirical fertigation management techniques with treatments in which advanced sensor­based fertiga­ Adv. Hort. Sci., 2023 37(1): 123­132 126 tion management was implemented. The main results of selected experiments carried out during the project are reported hereafter. The research activities focused on two model species [rocket salad (Diplotaxis tenuifolia L.) and tomato (Solanum lycopersicum L.) selected for their relevance in Mediterranean greenhouse vegetable production. In particular, rocket is reported as an e m e r g i n g l e a f v e g e t a b l e w h i c h c u l ti v a ti o n i s widespread and in further expansion (Schiattone et al., 2017), while tomato is the most important green­ house crop grown in soilless cultivation systems (Montesano et al., 2015). A study was carried out to test two irrigation scheduling approaches (timer­ or sensor­based) and two fertilization levels (high or low, with reference to the standard dosage range recommended for the specific fertilizers used) of open­cycle soilless rocket in Mediterranean autumn­winter unheated green­ house conditions (Montesano et al., 2021). Rocket plants (cv. Dallas, Isi Sementi) were grown in a peat:perlite (3:1) mixture in 4.5 L plastic pots. Four treatments were compared: timer with high or low fertilization (T­HF, T­LF), and sensor­based with high or low fertilization (S­HF, S­LF). In timer­based treat­ ments, irrigation schedule was periodically adjusted based on leaching fraction measurements (≈35% was set as a target, according to common practice). In sensor­based treatments, on­demand irrigation was operated based on substrate EC/temperature/mois­ ture sensors (GS3, Decagon Devices). These were connected to a CR1000 datalogger programmed to automatically open irrigation valves and supply water enough to constantly maintain volumetric water con­ tent to a pre­defined set­point (0.35 m3 m­3, close to maximum water holding capacity), with no leaching. Slow­release fertilizers (Osmocote Exact and CalMag, ICL) were mixed with the substrate at high (3.75 and 1 g L­1, respectively) or low dosage (2.25 and 0.6 g L­ 1). Yield, quality, water use and substrate parameters trends were evaluated. Sensors improved water use efficiency compared to timer (34.4 vs 21.4 g FW L­1, on average) matching water supply with plant needs, and preventing leaching (Fig. 1) (no interactive effects of fertilization treatments were observed on those parameters). Sensor­based irrigation also pro­ vided the best plant growth conditions, with interest­ ing interactive effects with fertilization rate. In partic­ ular, the highest and the lowest cumulative (three harvests) yield values were obtained in S­HF and T­LF respectively (144.8 and 102.2 g FW pot­1), while simi­ lar values were observed in S­LF and T­HF (131.4 g FW pot­1, on average) (Fig. 1). The partial fertilizer factor productivity (g product fresh weight / g fertiliz­ ers applied) was higher at low dosage, and, with the same dosage, when the sensors were used (Fig. 1). After each harvest time the fresh­cut rocket leaves were immediately transported in refrigerate condi­ tions to the postharvest laboratory (see WP2 section below) (Palumbo et al., 2021 b). In another set of studies, we aimed to apply approaches for the sustainable fertigation manage­ ment of soilless tomato (semi­closed cycle recircula­ tion; sensor­based nutrient solution supply manage­ ment) in comparison with a traditional open cycle free­drain nutrient solution management providing the use timer for fertigation schedule. Experiments were conducted with different tomato types (cherry ­ cv. Carminio, Seminis­Bayer, and intermediate type ­ cv. Mose, Syngenta), and in different environmental conditions typical of Mediterranean areas (including the use of brackish water for nutrient solution prepa­ ration). In general, both approaches (semiclosed­ cycle cultivation and open cycle with sensor­based fertigation management) reduced the environmental impact of the production process (reduced water/fer­ Fig. 1 ­ Water use efficiency (WUE), leaching rate, total yield, and partial fertilizer factor productivity of rocket (Diplotaxis tenuifolia) grown in open free drain soilless system with timer­ (T) or sensor­based (S) irrigation man­ agement, and subjected to high (HF) or low (LF) fertiliza­ tion rate. Amodio et al. ‐ The SUS&LOW project 127 tilizers usage; less nutrient solution released into the environment, increased water use efficiency) and positively affected tomato quality traits, compared to empirically management open­free drain cultivation. WP2: non‐destructive discrimination for low‐impact practices and non‐destructive quality assessment NIR spectroscopy and Hiperspectral imaging. In this WP, the objective of the tasks was to assess the potentiality of Fourier transformed­near infrared (FT­ NIR) spectrometry and hyperspectral imaging (HSI) to discriminate tomatoes and rocket leaves produced with different level of input as described in WP1, tak­ ing also into account the degree of efficiency in water and fertilizers used efficiency (WUE and FUE indexes). A hyperspectral line­scan scanner (Version 1.4, DV srl, Padova, Italy) equipped with two spectrographs, one in the Vis­NIR range, and the second in the NIR range, was used to obtain the HS images. The Vis­NIR spec­ trograph (400­1000 nm) has a spatial resolution of 1000 × 2000 pixels with a spectral resolution of 5 nm and was connected to a CCD camera. As for the NIR spectrograph (900­1700 nm), the spatial resolution was 600 × 320 pixels with a spectral resolution of 5 nm; and a CMOS (Specim Spectral Imaging Ltd., Oulu, Finland) with 50 frames per second equipped with C­ mount lenses was used. As for FT­NIR spectrometry an MPA Multi­Purpose (FT­NIR Analyzer, Bruker Optics, Ettlingen, Germany), was used during spectral acquisition over the range of 800­2777 nm (sphere macrosample re­solution 1.71 nm, scanner velocity 10 kHz, sample scan time 64 scans, background scan time 64 scans). After image processing and spectra extraction for the HSI, all spectra belonging to HSI and FT­NIR were tested in discrimination using the agronomic treatments as discriminant classes and Partial Least Squares­Discriminant Analysis (PLS­DA) as classification technique. As for rocket leaves, PLS­ DA was conducted with the 4 classes (T­HF; T­LF; S­ HF, S­LF) described in the paragraph related to WP1, using 70 percent of samples for calibration purpose and the remaining 30% for the external validation. The model performance was evaluated based on the accuracy, which is an average of the sensitivity calcu­ lated over the various classes, and gives an overall idea of the goodness of the classification. Results indicated HSI as a promising technique for the dis­ crimination of rocket produced with different cultural techniques, with an accuracy of classification in the prediction phase of 97.2% in Vis­NIR and 99.5% in NIR range. In figure 2, the results of the discrimination models can be observed. Regarding tomato, where 2 experiments with 2 different varieties were conducted (WP1), for each trial a first PLS­DA was aimed to discriminate the three treatments of cultivation and a second discrimi­ nation was performed for different levels of WUE and FUE. According to the efficiency of use of water and fertilizers we could individuate 2 levels (high and low) in each experiment and 3 levels (High, medium, and l o w ) m e r g i n g t h e d a t a o f b o t h e x p e r i m e n t s . Therefore, a PLS­DA with 3 levels of WUE (and FUE) was also generated with the full dataset. Among the different non­destructive techniques, FT­NIR and HIS in the VIS­NIR range gave comparable performances in discriminating tomato according to cultural prac­ tices and different use of sources. Discrimination for WUE for each variety improved the classification results, respect to the individual treatments, but the highest accuracy was obtained when the discrimina­ tion was based on 3 levels of WUE merging the 2 datasets, reaching 92.1%. In literature there are no studies aimed to discriminate crops for WUE or FUE, while we may find the application of HSI for the clas­ sification of water­stressed plants, as for the case of tomatoes (Rinaldi et al., 2015). In comparison to this study, reporting a mean accuracy of around 77% for discrimination of the two differently irrigated areas, our findings showed higher accuracy, exploring new area of the application for these techniques. Application of CVS for non‐destructive quality evalua‐ tion on packaged products A research activity was carried out to develop and validate an innovative CVS integrating a Random Forest model for classification: this model automati­ cally selects from the image the most relevant colour features for the task of interest. The developed CVS was applied to digital images of fresh­cut rocket Fig. 2 ­ Estimated class index values in the calibration and in the prediction process for the classification based on PLS­DA modes shown on table 2 in a) VNIR range (left) b) NIR range (right). f d d ld h 128 Adv. Hort. Sci., 2023 37(1): 123­132 leaves cultivated with LIP (WP1) to objectively esti­ mate the evolution of their quality levels (QL) during storage and to discriminate the cultivation approach applied on field. At harvest, rocket leaves were stored at 10°C in open polypropylene (PP) bags for a number of days required to reach the lowest QL, according to the rating scale from 5 (very good) to 1 (very poor), as reported in figure 3. At each QL, all the samples were subjected to postharvest quality evaluation, detecting colour parameters by a traditional colorimeter (CR400, Konica Minolta, Osaka, Japan) and physical and chem­ ical parameters, in detail respiration rate (Kader, 2002), electrolyte leakage (Kim et al., 2005) and total chlorophyll content (Cefola and Pace, 2015). Then, images of the same samples were acquired by the CVS for non­destructive quality assessment and for recognizing traits related to the sustainability of the cultivation management used on the field, with spe­ cific reference to water and nutrients use (WP1). Image pre­processing was applied: to separate the product from the background; to identify the colour­ chart placed in the scene to estimate the effects of lights and of the sensors and to correct colours to minimize these effects. Three colour correction meth­ ods (white balance, linear correction, and polynomial correction) with increasing level of complexity were evaluated and compared in terms of consistency of colour measurements and of classification perfor­ mance. Linear colour correction proved to be the best trade­off between efficacy and efficiency providing a slightly lower performance than polynomial correc­ tion with significantly simpler computation. Finally, a Random Forest model was used to train classifiers to assess the QL of rocket leaves and to identify the treatments used during the cultivation. All the postharvest quality parameters measured by traditional destructive methods were significant in QL assessment of fresh­cut rocket leaves. The pro­ posed classifier based on the Random Forest model was able to identify and select the most relevant colour traits for both the tasks (QL assessment and treatment identification) without human interven­ tion. The accuracy achieved in evaluating QLs of rock­ et leaves during storage was high (about 95%), while the performance in discriminating the cultivation approach was lower and not sufficient for practical applications (about 65­70%). Indeed, the different cultivation approaches did not significantly affect the visual characteristics of the product and the destruc­ tive measures: this task needs further investigations. Another research activity was carried out to develop and validate the capability of the non­destructive and contactless CVS to assess the visual quality changes during the cold storage of fresh­cut rocket leaves coming from soil and soilless growing systems (WP1) and to estimate some internal quality attributes (chlorophyll and ammonia content) also through the packaging material. Evaluating quality through the package is critical to identify the regions of the bag where the product is visible without shadows or highlights created by illumination: this is mandatory to measure colour properties in a reliable and mean­ ingful way. At harvest, rocket leaves, cultivated on soil or soilless system (WP1), were packed in open PP bags and stored at 10°C for about 18 d. During stor­ age, all samples were observed to attribute the QL according to the rating scale reported in figure 3 and the postharvest quality traits were evaluated by destructive conventional methods [colour parame­ ters, chlorophyll content, ammonia content (Fadda et al., 2016) and electrolyte leakage]. Then, images of unpackaged and packaged samples were acquired by the CVS. During image acquisition, no constraints were imposed on the position of the product in the bag, on the position of the bag in the scene or on the highlights created by the illumination on the surface of the bag: this was necessary to demonstrate the applicability of this technology into a real industrial line. Colour correction was performed by the linear model, identified as the best trade­off between effec­ tiveness and computational complexity in the previ­ ous research activity. Packed and unpacked products were processed using exactly the same phases apart from the artefacts’ elimination step applied to the images of packaged products to select the regions where the colour information was meaningful, with­ out interference from light artefacts and reflections. At last, the Random Forest model was used to solve Fig. 3 ­ Changes in the sensory quality level (QL) of fresh­cut rocket leaves during the storage at 10°C according to the 5 to 1 rating scale reported by Palumbo et al. (2021 b). In detail, QL5= very good; QL4= good; QL3= fair; QL2= poor; QL1= very poor. Amodio et al. ‐ The SUS&LOW project 129 both the classification problem (assessment of the QLs) and the regression problems (estimation of quality marker parameters such as chlorophyll and ammonia contents). The same architecture was used for all the tasks, by simply changing the training data. The histogram of the image, evaluated in the a­b plane of the CIELab colour space, was used as the set of features. The Random Forest model was able to automatically select the subset of values more suit­ able for solving each task. All the postharvest quality parameters detected by conventional analysis during the storage of fresh­ cut rocket leaves were significant in the QL assess­ ment and, among them, chlorophyll and ammonia contents proved to be useful marker parameters for the objective separation of each QL considered, both on soil and soilless cultivation approach. The CVS was able to operate without relevant dif­ ferences on unpackaged and packaged products. The test was done joining all the samples, regardless of the cultivation approach: the results showed a not significant performance loss on packaged leaves (Pearson’s linear correlation coefficient of 0.84 for chlorophyll and 0.91 for ammonia) with respect to unpackaged ones (0.86 for chlorophyll and 0.92 for ammonia) (Fig. 4). Finally, three Partial Least Square (PLS) models were performed to predict the QL using as predictors chlorophyll and ammonia contents obtained by destructive methods (Model I), by CVS on packaged products (Model II) and by CVS on unpackaged ones (Model III) (Table 1). The results showed high performances in terms of R2 and the model obtained by predictors estimated non­destructively by the CVS (Model II and III) provid­ ed better performances in the QL prediction than one obtained by destructive analysis, in both calibration and validation. WP3: marketing strategies to support the added value of the products LIP and ND certified Implementing a marketing strategy, based on often intangible characteristics to consumers such as LIP and ND, it is not an easy task. Low impact prac­ tices do not have a highly distinctive impact on prod­ uct characteristics nor determine unique taste, flavour, or look elements to consumers. However, certifications could be used to signal quality through the application of standards of quality and practices. Whether certifications could be effective in terms of marketing in the case of products LIP and ND, or for signalling quality in general is matter of discussion. Vecchio and Annunziata (2011), for instance, in their work question the possibility of effective understand­ ing of certification by consumers. At this purpose the research team of WP3 decided to implement a differ­ ent strategy and test it on the market. Visual story­ telling certifying LIP and ND has been then hypothe­ sized to better communicate the importance and the impact of those practices on food. The research activity, therefore, has been orga­ nized in three steps: identifying the communication Fig. 4 ­ Values estimated by the CVS (abscissa) vs. values mea­ sured in the laboratory (ordinate) for ammonia content on unpackaged (A) and packaged (B) rocket leaves and for total chlorophyll content on unpackaged (C) and packaged (D) samples (Palumbo et al., 2022 b). Table 1 ­ Root Mean Square Error (RMSE) and the coefficient of determination (R2) in calibration (c) or validation (v) of the Partial Least Square (PLS) Models predicting visual quality of rocket leaves (Palumbo et al., 2022 b) PLS Models Predictors RMSEc R 2 c RMSEv R 2 v I Total chlorophyll and ammonia obtained by destructive methods 0.45 0.9 0.86 0.70 II Total chlorophyll and ammonia obtained by CVS on packaged rocket leaves 0.46 0.89 0.75 0.77 III Total chlorophyll and ammonia obtained by CVS on unpackaged rocket leaves 0.46 0.89 0.7 0.8 Adv. Hort. Sci., 2023 37(1): 123­132 130 strategy and set­up; testing through focus­groups the opportunity conditions for farms and companies; testing though a survey and an econometric analysis the consumers’ preference and their willingness to pay for products with LIP and ND. Therefore, a draft platform has been developed containing basic com­ munication rules in order to highlight sustainability attributes of products through storytelling. Workflow has been established and a simulation has been con­ ducted (Fig. 5). Focus group with producers has allowed verifying the general appreciation for the marketing approach and allowed a better set­up of the strategy. Finally, a picture­based simulation has been produced for the final test and the survey to consumers (Fig. 6). As last activity, a questionnaire based survey has been prepared and administered to 467 consumers and an econometric model to estimate willingness to pay and consumers orientation has been set up and then estimated. The whole set of activties within the research project allowed understanding how impor­ tant is a correct communication of products and how different could be the perception of a product based o n h o w y o u c e r ti f y o r n a r r a t e t h e p r o d u c ti o n method. Result allow understanding that older con­ umsers are more aware of sustainability and are more willing to pay for LIP products. Psicological pro­ file such as traditionalism and benevolence identify the consumer that, more than other profiles, would be willing to pay a higher price. 3. Conclusions Sensor­based fertigation management applied to rocket leaves and tomato confirmed to be a feasible approach to improve sustainability of soilless cultiva­ tion, also in cases where the complete and rapid switch to closed cycle recirculation systems is still impaired by economic, social, and environmental fac­ tors such as in Mediterranean area. The results of this project related to non­destruc­ tive discrimination of tomatoes and rocket leaves, according to cultural practices using different levels of inputs (water and fertilizers), indicated the high potentiality of HSI and FT­NIR techniques for the authentication of sustainable growing methods. Moreover, project activities’ proved CVS as an effec­ tive tool for evaluating the product quality also through the bag, even working only on the regions of the image that provide meaningful colour information about the product’s surface. The integration of machine learning modules inside the CVS confirmed to be useful to simplify the design and tuning, done mostly automatically without human intervention. Moreover, the flexibility introduced by machine learning makes the resulting architecture more flexi­ ble in adapting to different products and applications. As regards the marketing approach, consumers resulted willing to pay a higher price for LIP products certified through a visual storytelling platform. In the next future, there could be a good chance that sus­ tainability­oriented practices coupled with a visual storytelling certification style could gain shares on food markets. Acknowledgements This research was funded by the project Prin 2017 “SUS&LOW­Sustaining low­impact practices in horti­ culture through non­destructive approach to provide more information on fresh produce history and quali­ ty” (grant number: 201785Z5H9) from the Italian Ministry of Education University. References AMODIO M.L., CABEZAS­SERRANO A.B., PERI G., COLELLI G., 2011 ­ Post‐cutting quality changes of fresh‐cut arti‐ chokes treated with different anti‐browning agents as evaluated by image analysis. ­ Postharvest Biol. Technol., 62: 213­220. Fig. 5 ­ Workflow for products LIP and ND certified platform. Fig. 6 ­ Picture based simulation of visual storytelling certifica­ tion for LIP producs and ND. Amodio et al. ‐ The SUS&LOW project 131 AMODIO M.L., CAPOTORTO I., CHAUDHRY M.M.A., COLEL­ LI G., 2017 a ­ The use of hyperspectral imaging to pre‐ dict the distribution of internal constituents and to clas‐ sify edible fennel heads based on the harvest time. ­ Comput. Electron. Agric., 134: 1­10. AMODIO M.L., CHAUDHRY M.M.A., COLELLI G., 2017 b ­ The use of non‐destructive techniques to assess the nutritional content of fruits and vegetable, pp. 763­780. ­ I n : Y A H I A E . M . ( e d . ) F r u i t a n d v e g e t a b l e Phytochemicals: Chemistry and human health. Wiley­ Blackwell, London, UK, pp. 1488. AMODIO M.L., CHAUDHRY M.M.A., COLELLI G., 2020 ­ Spectral and hyperspectral technologies as an addition‐ al tool to increase information on quality and origin of horticultural crops. ­ Agronomy, 10(1): 7: 1­10. ARENDSE E., FAWOLE O.A., MAGWAZA L.S., OPARA U.L., 2018 ­ Non‐destructive prediction of internal and exter‐ nal quality attributes of fruit with thick rind: A review. ­ J. Food Eng., 217: 11­23. ARIANA D.P., LU R.F., 2010 ­ Evaluation of internal defect and surface color of whole pickles using hyperspectral imaging. ­ J. Food Eng., 96(4): 583­590. A R I A S R . , L E E T . C . , L O G E N D R A L . , J A N E S H . , 2 0 0 0 ­ Correlation of lycopene measured by HPLC with the L*, a*, b* color readings of a hydroponic tomato and the relationship of maturity with color and lycopene con‐ tent. ­ J. Agric. Food Chem., 48(5): 1697­1702. BABELLAHI F., PALIWAL J., ERKINBAEV C., AMODIO M.L., CHAUDHRY M.M.A., COLELLI G., 2020 ­ Early detection of chilling injury in green bell peppers by hyperspectral i m a g i n g a n d c h e m o m e t r i c s . ­ P o s t h a r v e s t B i o l . Technol., 111100. BONA E., MARQUETTI I., LINK J.V., MAKIMORI G.Y.F., DA COSTA ARCA V., LEMES A.L.G., FERREIRA J.M.G., SCHOLZ M.B.S., VALDERRAMA P., POPPI R.J., 2017 ­ Support vector machines in tandem with infrared spec‐ troscopy for geographical classification of green arabi‐ ca coffee. ­ LWT ­ Food Sci. Technol., 76: 330­336. CAVALLO D.P., CEFOLA M., PACE B., LOGRIECO A.F., ATTOLICO G., 2018 ­ Non‐destructive automatic quality evaluation of fresh‐cut iceberg lettuce through packag‐ ing material. ­ J. Food Eng., 223: 46­52. CAVALLO D.P., CEFOLA M., PACE B., LOGRIECO A.F., ATTOLICO G., 2017 ­ Contactless and non‐destructive chlorophyll content prediction by random forest regres‐ sion: A case study on fresh‐cut rocket leaves. ­ Comput. Electron. Agric., 140: 303­310. CEFOLA M., PACE B., 2015 ­ Application of oxalic acid to preserve the overall quality of rocket and baby spinach leaves during storage. ­ J. Food Process. Preserv., 39: 2523­2532. FADDA A., PACE B., ANGIONI A., BARBERIS A., CEFOLA M., 2016 ­ Suitability for ready‐to‐eat processing and preservation of six green and red baby leaves cultivars and evaluation of their antioxidant value during stor‐ age and after the expiration date. ­ J. Food Process. Preserv., 40(3): 550­558. FAO, 2013 ­ Good agricultural practices for greenhouse vegetable crops. Principles for Mediterranean climate areas. ­ FAO, Food and Agriculture Organization of the United Nation. Plant Production and Protection, Rome, Italy, Paper 217. GUO Z., HUANG W., CHEN L., ZHAO C., PENG Y., 2013 ­ Geographical classification of apple based on hyper‐ spectral imaging. ­ In: MOON S.K., T. SHU­I, and C. KUANGLIN (eds.) Sensing for agriculture and food quali‐ ty and safety V, SPIE, 8721: 103­110. KADER A.A., 2002 ‐ Methods of gas mixing, sampling and a n a l y s i s , p p . 1 4 5 ­ 1 4 8 . ‐ I n : K A D E R A . A . ( e d . ) P o s t h a r v e s t t e c h n o l o g y o f h o r ti c u l t u r a l c r o p s . U n i v e r s i t y o f C a l i f o r n i a A g r i c u l t u r e a n d N a t u r a l Resources, Oakland, CA, USA, pp. 535. KHANMOHAMMADI M., KARAMI F., MIR­MARQUÉS A., GARMARUDI A.B., GARRIGUES S., DE LA GUARDIA M., 2014 ­ Classification of persimmon fruit origin by near infrared spectrometry and least squares‐support vector machines. ­ J. Food Eng., 142: 17­22. KIM J.G., LUO Y., TAO Y., SAFTNER R.A., GROSS K.C., 2005 ­ Effect of initial oxygen concentration and film oxygen transmission rate on the quality of fresh‐cut romaine lettuce. ‐ J. Sci. Food Agric., 85: 1622­1630. KRYSTALLIS A., CHRYSSOHOIDIS G., 2005 ­ Consumers’ will‐ ingness to pay for organic food: Factors that affect it and variation per organic product type. ­ Br. Food J., 107(5): 320­343. LARA M.A., LLEÓ L., DIEZMA­IGLESIAS B., ROGER J.M., RUIZ­ALTISENT M., 2013 ­ Monitoring spinach shelf‐life with hyperspectral image through packaging films. ­ J. Food Eng., 119(2): 353­361. LIU Y., PU H., SUN D.W., 2017 ­ Hyperspectral imaging technique for evaluating food quality and safety during various processes: A review of recent applications. ­ Trends Food Sci. Technol., 69: 25­35. MASSA D., MAGÁN J.J., MONTESANO F.F., TZORTZAKIS N., 2020 ­ Minimizing water and nutrient losses from soil‐ less cropping in southern Europe. ­ Agric. Water Manage., 241, 106395. MONTESANO F.F., D’IMPERIO M., RENNA M., PARENTE A., LEONI B., SERIO F., 2021 ­ Soilless production of rocket (Diplotaxis tenuifolia L.) as affected by low and high input of water and fertilizers. Poster presentation at “III International Symposium on Soilless Culture and Hydroponics: Innovation and Advanced Technology for Circular Horticulture”, ISHS Hydro2020, Lemesos, Cyprus, 19­22 March. MONTESANO F.F., SERIO F., MININNI C., SIGNORE A., PAR­ ENTE A., SANTAMARIA P., 2015 ­ Tensiometer‐based irrigation management of subirrigated soilless tomato: effects of substrate matric potential control on crop performance. ­ Front. Plant Sci., 6: 1150. MONTESANO F.F., VAN IERSEL M.W., BOARI F., CANTORE V., D’AMATO G., PARENTE A., 2018 ­ Sensor‐based irri‐ https://www.sciencedirect.com/science/article/pii/S0023643816302328#! https://www.sciencedirect.com/science/article/pii/S0023643816302328" /l "! https://spie.org/profile/Moon.Kim-29716 https://spie.org/profile/Kuanglin.Chao-24021 Adv. Hort. Sci., 2023 37(1): 123­132 132 gation management of soilless basil using a new smart irrigation system: Effects of setpoint on plant physiolog‐ ical responses and crop performance. ­ Agric. Water Manage., 203: 20­29. PACE B., CAVALLO D.P., CEFOLA M., COLELLA R., ATTOLICO G., 2015 ­ Adaptive self‐configuring computer vision system for quality evaluation of fresh‐cut radicchio. ­ Innov. Food Sci. Emerg. Technol., 32: 200­207. PACE B., CEFOLA M., DA PELO P., RENNA F., ATTOLICO G., 2014 ­ Non‐destructive evaluation of quality and ammonia content in whole and fresh‐cut lettuce by computer vision system. ­ Food Res. Int., 64: 647­655. PACE B., CEFOLA M., RENNA F., ATTOLICO G., 2011 ­ Relationship between visual appearance and browning as evaluated by image analysis and chemical traits in fresh‐cut nectarines. ­ Postharvest Biol. Technol., 61(2): 178­183. PACE B., CEFOLA M., RENNA F., RENNA M., SERIO F., ATTOLICO G., 2013 ­ Multiple regression models and computer vision systems to predict antioxidant activity and total phenols in pigmented carrots. ­ J. Food Eng., 117(1): 74­81. PALUMBO M., ATTOLICO G., CAPOZZI V., COZZOLINO R., CORVINO A., DE CHIARA M.L.V., PACE B., PELOSI S., R I C C I I . , R O M A N I E L L O R . , C E F O L A M . , 2 0 2 2 a ­ Emerging postharvest technologies to enhance the shelf‐life of fruit and vegetables: An overview. ­ Foods, 11(23): 3925. PALUMBO M., D’IMPERIO M., TUCCI V., CEFOLA M., PACE B., SANTAMARIA P., PARENTE A., MONTESANO F.F., 2021 a ­ Sensor‐based irrigation reduces water con‐ sumption without compromising yield and postharvest quality of soilless green bean. ­ Agronomy, 11: 2485. PALUMBO M., PACE B., CEFOLA M., MONTESANO F.F., COLELLI G., ATTOLICO G., 2022 b ­ Non‐destructive and contactless estimation of chlorophyll and ammonia contents in packaged fresh‐cut rocket leaves by a Computer Vision System. ­ Postharvest Biol. Technol., 189: 111910. PALUMBO M., PACE B., CEFOLA M., MONTESANO F.F., SERIO F., COLELLI G., ATTOLICO G., 2021 b ‐ Self‐config‐ uring CVS to discriminate rocket leaves according to cultivation practices and to correctly attribute visual quality level. ­ Agronomy, 11(7): 1353. PATEL K.K., KAR A., JHA S.N., KHAN M.A., 2012 ­ Machine vision system: a tool for quality inspection of food and agricultural products. ­ J. Food Sci. Technol., 49(2): 123­ 141. PIAZZOLLA F., AMODIO M.L., COLELLI G., 2013 ­ The use of hyperspectral imaging in the visible and near infrared region to discriminate between table grapes harvested at different times. ­ J. Agric. Eng., 44(2): 49­55. PIAZZOLLA F., AMODIO, M.L., COLELLI, G., 2017 ­ Spectra evolution over on‐vine holding of Italia table grapes: prediction of maturity and discrimination for harvest times using a Vis‐NIR hyperspectral device. ‐ J. Agric. Eng., 48(2): 109­116. RENNA M., D’IMPERIO M., MAGGI S., SERIO, F., 2022 ­ Soilless biofortification, bioaccessibility and bioavail‐ ability: signposts on the path to personalized nutrition. ­ Front. Nutr., 9: 966018. RINALDI M., CASTRIGNANÒ A., DE BENEDETTO D., SOLLIT­ TO D., RUGGIERI S., GAROFALO P., SANTORO F., FIGO­ R I T O B . , G U A L A N O S . , T A M B O R R I N O R . , 2 0 1 5 ­ Discrimination of tomato plants under different irriga‐ tion regimes: analysis of hyperspectral sensor data. ‐ Environmetrics, 26(2): 77­88. ROUPHAEL Y., KYRIACOU M.C., 2018 ­ Enhancing quality of fresh vegetables through salinity eustress and bioforti‐ fication applications facilitated by soilless cultivation. ­ Front. Plant Sci., 9: 1254. SÁNCHEZ M.T., GARRIDO­VARO A., GUERRERO J.E., PÉREZ­ MARÍN D., 2013 ­ NIRS technology for fast authentica‐ tion of green asparagus grown under organic and con‐ ventional production systems. ­ Postharvest Biol. Technol., 85: 116­123. SCHIATTONE M.I., CANDIDO V., CANTORE V., MONTESANO F.F., BOARI F., 2017 ­ Water use and crop performance of two wild rocket genotypes under salinity conditions. ­ Agric. Water Manage., 194: 214­221. STANGHELLINI C., 2014 ­ Horticultural production in green‐ houses: Efficient use of water. ­ Acta Horticulturae, 1034: 25­32. TORRELLAS M., ANTÓN A., LÓPEZ J.C., BAEZA E.J., PARRA J.P., MUÑOZ P., MONTERO J.I., 2012 ­ LCA of a tomato crop in a multi‐tunnel greenhouse in Almeria. ­ Int. J. Life Cycle Assess., 17(7): 863­875. TSOUVALTZIS P., BABELLAHI F., AMODIO M.L., COLELLI G., 2020 ­ Early detection of eggplant fruit stored at chill‐ ing temperature using different non‐destructive optical techniques and supervised classification algorithms. ­ Postharvest Biol. Technol., 159: 1­9. VECCHIO R., ANNUNZIATA A., 2011 ­ The role of PDO/PGI labelling in Italian consumers’ food choices. ­ Agric. Econ. Res. Rev., 12(2): 80­98. WORKMAN J., SHENK J., 2004 ­ Agronomy Monograph. ­ Near­Infrared Spectroscopy in Agriculture, 44:3­10. YANG Y.C., SUN D.W., PU H., WANG N.N., ZHU Z., 2015 ­ Rapid detection of anthocyanin content in lychee peri‐ carp during storage using hyperspectral imagine cou‐ pled with model fusion. ­ Postharvest Biol. Technol., 103: 55­65.