For instance, 75 percent of the articles returned in our Web of Science search were published after 2009 and 18 percent were published in 2017-18. A more recent search of these terms in April of 2018 returns 1622 records revealing a continued growth in literature on urban agriculture. Of these records, journal articles dominate . Other records include book reviews, article reviews, proceedings papers, and meeting abstracts. The main contributing journals included Land Use Policy , Landscape & Urban Planning , Agriculture & Human Values , Sustainability , and Local Environment ; however, the sources were quite diverse. Each record represents a single document and together they form the corpus used to build the reference topic model. Prior to processing, we removed any stop words, punctuation, and URLs. We performed LDA topic modelling using the MALLET program . We produced various topic models using three granularities , and used the models with the greatest log likelihood . We then examined their topic composition and removed topics dominated by non-meaning-bearing terms including time and location indicators and general publication information . These topics were identified using the alpha hyperparameter, where relatively high values indicated that the topic was common throughout the corpus and therefore not meaningful for examining differences within our sample. After these adjustments, we determined that the 25-topic model was ideal for analysis using personal expert knowledge on urban agriculture literature. The reference topic model was created in order to perform inference on content produced by urban agriculture growing sites and regional organizations in San Diego County – in other words, to interpret the content produced by the key actors identified above . We created a corpus including all textual content from the websites of agencies in our sample,plant pot with drainage with content from each of the 48 observations contained in a single document in the corpus.
Textual content included any written descriptions on the website including history, mission and vision statements, program descriptions, excluding locations, contact, and event info. For growing sites associated with larger organizations or institutions, we also collected basic descriptive content from the parent website. By applying the reference model to all the documents, each document is characterized in terms of topic composition, allowing comparisons among documents . The output of the inferencing process is a document-topic distribution matrix, from which we computed a matrix of cosine similarities among documents. In order to visualize these similarities, we used a dimensionality reduction technique known as multidimensional scaling . In the resulting output, each document is described as a 2-D point in Cartesian coordinates, where proximity relates to similarity. The resulting discursive map displayed the inferenced website corpus, with each point representing a single growing site or organization. The location of each point relates to its particular topic composition. The distance between points is indicative of their discursive similarity – the closer two points are in the discursive map, the more similar their topic composition; the farther apart, the more dissimilar. We investigated this map, but also created a series of variations, altering the symbology of the discursive map to reflect particular features of the sites. This allowed us to examine the connections between characteristics like growing methods and topic composition. We were also interested in discovering clusters among the data points, and so we utilized k-means clustering to identify meaningful groups in our data . K-means is a heuristic algorithm that attempts to partition aninput dataset into k groups, allowing researchers to explore clusters within a dataset. Our data seemed to occupy primarily three quadrants in the discursive map, and so we chose to identify three classes. This algorithm was run for 1,000 iterations and the results with the lowest sum of squared errors – a metric that explains the difference between each observation and its corresponding k-means centroid – were chosen as representative. This analysis complemented our visual analysis of symbology patterns. The growing methods symbology illustrating the practices used by growing sites revealed a distinct, but blurry pattern between motivation and practice.
When analyzing the map using this symbology, a general pattern emerged in which technologically-advanced sites tended to group in the top-left quadrant of the map with two outliers: Go Green Agriculture and Archi’s Acres. The absence of innovation in these outliers’ top-three loadings suggested that other topics precede technology in how these growing sites describe themselves despite their use of advanced technologies. Generally, soil-based sites occupied the right side of the discursive map; however, soil-based farms such as Suzie’s Farm, Good Taste Farm, and Point Loma Farms were grouped in with the soilless sites. Growing site and organization descriptions of their processes did drive their location on the discursive map. For instance, the soilless sites often described the inventive and underrepresented practices they use to grow produce in the urban environment. However, the content did not end there. Other topics like social movements, climate change, and food access were also present among these sites. We saw a similar trend with sites using a community gardening model. When we explored the entire topic loadings of growing sites and organizations, ignoring practice-based topics like innovation and community gardening topics, we saw that the clusters have far more similarities than differences. Interestingly, these soilless sites are typically affiliated with businesses as opposed to nonprofits which dominate the right side of the map, where most soil-based sites are located . Indeed, we expected that business and nonprofit website content would vary and these results provide evidence to that effect. San Diego Food System Alliance, the leading regional nonprofit organization, is located in the center of the map. This location is not surprising in the context of neoliberal governance in which cities and regional organizations are more focused on building consensus and supporting apolitical agendas, rather than taking on political causes .The affiliation symbology illustrating the relationship between institutional affiliation and content was less coherent than the other symbologies displayed in previous figures, but still offered important insights. Growing sites were affiliated with a variety of institutions including schools, churches, organizations hosting training and educational programs, and for-profit businesses.
Education sites were located throughout the map suggesting that training and skill-building are not major dividing factors in discourse. In other words, many different types of organizations claim to focus on education. However, church, community, and school gardens tended to concentrate in the top-right section of the map, which is typically associated with soil-based community gardens. However, it cannot be assumed that the for-profit sites lack social mission. For example, Archi’s Acres, a for-profit hydroponic farm in Escondido, includes a social enterprise function focusing on training veterans in hydroponic farming. Sundial Farms, a veteran- and immigrant-owned, hydroponic farm in the Innovation cluster, is a direct result of this program. This social function features prominently in its website content: “At Archi’s, we believe a key aspect of successful business is how it meets its responsibility to the community in which it operates and the customers which make up its marketplace. We do this by integrating into our business model an opportunity to support others including our military service members and veterans.” This broader social mission may explain its topic loadings and the absence of innovation as a primary topic. The overall uniqueness of this growing site may explain its peripheral location in the discursive map. Solutions Farms, an aquaponic operation associated with Solutions for Change,growing blueberries in pots was the only nonprofit located in the for-profit dominated section of the map. The organization aims to alleviate family homelessness in the county through skill development, including training in aquaponic farming. However, innovation is the primary topic in their content, influencing their location among other sites whose discourse is focused on innovation.Multivariate clustering was performed on the discursive map to identify clusters in the sites and group them accordingly. Figure 8 contains the k-means results including three classes . Transitional sites were identified by creating a 4-class result . The topic compositions of sites in each cluster were examined and the clusters were given descriptive names reflecting their dominant topics : Innovation, Community, and Access. The transitional sites – those that broke off into their own group in the 4-class result – were signified using an overlaid line pattern. These sites were close to or straddled the center axes of the map.The Innovation cluster was distinct from the other clusters. The predominant topic amongst this group was innovation, which includes words and phrases like rooftop farming, zero-acreage farming, soilless, aquaponics, buildings, hydroponic, vertical, greenhouses, indoor, and technology as well as production, yield, growth, and quality. Unsurprisingly, all of the technologically-advanced sites resided in this cluster with the exception of Valley View Farms, which experiments with hydroponics, but focuses primarily on animal farming. Among the topic loadings in this group were community gardening, food access, social movements, climate change, water management, food production, and food security. This cluster also consisted primarily of for-profit growing sites with the exception of Roger’s Community Garden located on the University of California, San Diego campus. An interesting outlier is Go Green Agriculture, a hydroponic farm, which is located on the border of the Community cluster. This location is likely driven by its top topics, which include community gardening, location, and climate change, which are well-represented in both the Innovation and Community cluster. The Community cluster emphasized connections with local residents, primarily promoting home and community gardening – community gardening was the most prevalent topic in this cluster.
Although, this cluster overlaped considerably with the Access cluster, there was a clear emphasis on environmental topics including ecosystem conservation, water management, location, water contamination, innovation, and climate change. The social movement topic was also prevalent throughout this cluster with many of its sites expressing a dedication to alternative forms of organization. It is also worth noting that the socio-economic characteristics of the two neighborhoods are also quite different. Southeastern San Diego, specifically zip-code 92102 where Mt. Hope Community Garden is located, is a primarily Hispanic community, followed by White , African American , and Asian . The median income is at $42,464 with only 24 percent of the population exceeding $75,000 annually . The sites and organizations in this cluster also placed considerably less emphasis on environmental topics in favor of more social topics including public health, food production, and urban greening. Still, topics like ecology and climate change were present suggesting that environmental and social concerns were not mutually exclusive. The sites in the Access cluster were also predominantly affiliated with educational and training programs. Two particularly interesting examples are UrbanLife Farms and Second Chance Youth Garden. Both growing sites are wings of social justice organizations that offer job training and skills development for youth living in City Heights and Southeastern San Diego – communities that have seen considerable disinvestment and suffer from high unemployment . Other growing sites like Rolling Hills Grammar School and Literacy Garden and Olivewood Gardens and Learning Center also focus on youth programming. Not all the growing sites in this cluster work with youth. New Roots Farm concentrates on providing resettled refugees with land for farming, small-business training, and nutrition education to help them adjust to a new life away from their home country. This mission guided its topic loading of food security, social movements, and food access. The five urban agriculture supporting organizations we surveyed spanned the Community and Access clusters. Slow Food San Diego, Slow Food Urban San Diego, San Diego Roots Sustainable Food Project , and San Diego Community Garden Network are located in the Community cluster. San Diego Food System Alliance was located at the border between the Community and Access clusters suggesting that food access was a more prominent topic for the organization. Further, its central position illustrated the consensus focus of the organization, which caters to a diverse group of actors including politicians, businesses, and nonprofit organizations. Overall, the placement of the organizations made sense as they are nonprofit facilitators for other sites aimed at broader social goals like increasing food access and building community.