Agricultural big data and methods and models for food security analysis—a mini-review

Survey methodology

Many authors have emphasized the importance of conducting a literature review because they consider it a valuable and qualified source of information that summarizes and adds to the body of knowledge in a particular field of study (Denyer & Tranfield, 2009; Knopf, 2006; Tranfield, Denyer & Smart, 2003). The best literature review should reveal, assess, and structure the relevant literature on the intended topic, as well as combine it with a critical analysis of various arguments in the literature (Denyer, Tranfield & Van Aken, 2008; Tranfield, Denyer & Smart, 2003). The goal of this study was to conduct a literature review to identify the global big data, methods, and models for FS and investigate how they can be used to improve FS levels globally, as well as in UAE (Denyer, Tranfield & Van Aken, 2008; Tranfield, Denyer & Smart, 2003). The least level of bias was ensured via a comprehensive literature inspection of the available published studies to provide an audit pathway from the decisions of the reviewers to the actions and conclusions (Munn et al., 2018; Tranfield, Denyer & Smart, 2003). Furthermore, selecting the research methodology required identifying, analyzing, and synthesizing the selected secondary data sources related to FS big data, methods, and models across a wide range of contexts and disciplines to provide a comprehensive understanding based on the fit to the review’s specified questions. According to (Denyer & Tranfield, 2009; Munn et al., 2018; Tranfield, Denyer & Smart, 2003), producing good and comprehensive systematic reviews is crucial for driving research, developing new research baselines, and opening multiple pathways for future research. As a result, a systematic literature review research method was selected to achieve the research objectives. Based on the approaches described by Denyer & Tranfield (2009), Munn et al. (2018), Tranfield, Denyer & Smart (2003), we conducted the review through five steps to ensure replicability and transparency, as detailed in Fig. 1. The research began with the formulation of a research questions with specific characteristics, such as being purposeful and specific. The scope and focus of the review were then defined. The goal of this study was to conduct a systematic literature review to identify the big data, methods, and models of FS in the global and UAE contexts. To answer the main research questions, this article provides a critical review of the existing literature published in Scopus and Web of Science databases (Martín-Martín et al., 2018). The following topics have been explored in this review of literature: (1) data extraction tools, (2) data format and infrastructure, (3) potential and limitations of agricultural big data (AgBD), and (4) FS methods and models. Thus, this article provides a reference for policymakers and practitioners, as well as a roadmap for future research, by highlighting the concerns in the areas mentioned above.

The second step involved creating a specific research criterion to ensure that the research sources chosen were sufficient and comprehensive enough to capture all the major points that adequately answer the research questions (Denyer & Tranfield, 2009). The necessity of understanding big data for FS in both the global and UAE contexts, with a strong emphasis on avoiding any source of bias during the selection process, was the key research gap that drove this study. As a result, the databases Scopus and Web of Science were used (Martín-Martín et al., 2018). Big data, methods, models, FS availability, FS access, FS stability, FS stability, and food infrastructure were among the keywords used. The keywords were chosen after a thorough examination of the most relevant concepts in the literature that affect each of the four FS dimensions. In July 2021, the research sources were chosen, and the title, abstract, and full-text searches for keywords were enabled. To find the available literature, several keywords were identified (Cooper et al., 2018). Primary and secondary keywords were used in the search strings. The purpose of using multiple strings was to cover as many articles as possible that dealt with the topic of FS or any of its four dimensions. The review was then subjected to specific exclusion and inclusion criteria to produce high-quality evidence (Tranfield, Denyer & Smart, 2003). To ensure that the review has a high quality, a reasonable number of articles were selected for in-depth analysis based on a set of exclusion and inclusion criteria (Fig. 1). Within the time frame (2010–2021), only peer-reviewed journal articles written in English were included in the review. Strict selection criteria were applied to the first search pool to maintain research transparency and ensure the selection of relevant material that answers the research questions (Kelly, Sadeghieh & Adeli, 2014; Xiao & Watson, 2017). After removing duplicated articles from both databases, a total of 130 articles were chosen for the review.

The fourth step entailed analyzing the selected 130 articles individually, summarizing and listing all big data, methods, and models for FS analysis, then synthesizing the extracted information from all sources to create new knowledge (framework), listing the similarities between all resources, and extracting the major insights globally and within the UAE context (Denyer & Tranfield, 2009; Tranfield, Denyer & Smart, 2003). The models and methods for FS analysis in each of the 130 articles were summarized using Microsoft Excel. The aggregative approach was then used for synthesis. The findings section includes a detailed report of answers to the following research questions: (1) data extraction tools, (2) data format and infrastructure, (3) potential and limitations of AgBD, and (4) FS methods and models. After that, the synthesis process was used to create a comprehensive framework that models big data and FS methods and models.

Data extraction tools

Data format and infrastructure

FS data can be unstructured or structured and stored in various formats, including TXT, JSON, and CSV. For instance, Singh, Shukla & Mishra (2018) collected SM data from Twitter in JSON and TXT formats, and then implemented the parsing method to covert JASON data to CSV data. There are also various formats for big data, such as raster and vector formats (SHP, TIF, CN, and NetcDF) (Ghiringhelli et al., 2017; Limbachiya & Gupta, 2015). On the other hand, Song et al. (2020) stored the data in relational databases with different attributes as a list of rows. Alfian, Syafrudin & Rhee (2017) used NoSQL and SQL databases to store IoT-generated sensor data with a large unstructured format and continuous data-generation characteristics. They also developed a real-time food quality monitoring system that employs sensor data from a smartphone and stores it in the MongoDB database.

Supercomputing and cloud computing are two major components of data infrastructure (Yang et al., 2017b). To address the challenges associated with big data, supercomputing must be considered. The United States has long been committed to supercomputing to facilitate knowledge exchange between the Exascale Computing Project and the industrial user community (Witze, 2014). The development of supercomputing infrastructures is also a priority for the EU. So far, the EU has built eight supercomputing centers to enhance bioengineering applications (Butler, 1999). Tianjin, Jinan, Changsha, Shenzhen, Guangzhou, Wuxi, and Zhengzhou are China’s seven national supercomputing centers. The Chinese government has created a food safety traceability platform (Berti & Semprebon, 2018) that collects 31 provincial food traceability data and connects national supercomputing centers. Its goal is to achieve food traceability from farm to fork while also providing services to food producers, such as food traceability, security, and oversight.

To enable big data research, several cloud computing infrastructures need to be developed (Yang et al., 2017a). In 2019, the EU Food Nutrition Security Cloud project (https://cordis.europa.eu/project/id/863059) aimed to integrate European research infrastructure by bringing together FNS data to address diet, health, and consumer behavior, as well as sustainable agriculture and bioeconomy. The Guizhou Food and Drug Administration in China released the food safety cloud system in 2014. It has now been transformed into an intelligent food safety supervision system, an internet plus inspection system, a traceability certification system, and a big data platform for government enterprises, testing institutions, and other social age organizations (Tao et al., 2018). Despite the importance of supercomputing and cloud computing infrastructures as distinct big data environments, the Middle East and North Africa (MENA) and GCC regions remain uninterested, necessitating significant attention to assist them in addressing food insecurity.

Agricultural big data (potential, current status, and limitations)

To meet the demands of the rapidly growing population, which is expected to reach nine billion people by 2050 (World Population Prospects, 2015), agricultural production and FSCs must be optimized by producing and delivering efficient food, feed, fiber, and fuel (Abe, 2017; Asseng et al., 2018; Asseng et al., 2019). This goal has become more difficult to achieve due to urbanization, climate change (Ali et al., 2022; Ali et al., 2020; Ding et al., 2021; Kheir et al., 2019), and water scarcity (Kheir et al., 2021b). AgBD will be a key component of the second green revolution, which will be required to meet the demands of the growing population. Furthermore, the crop growth simulation modeling approach has been proven to be a useful tool for determining the impact of climate uncertainty on crop yields (Asseng et al., 2013; Ejaz et al., 2022; Shoaib et al., 2021). Many countries and commodity markets are already using AgBD to detect supply chain disruptions in commodity crops like wheat, rice, corn, and soybean (Bock & Kirkendall, 2017; George Hanuschak, 1993; Rosenzweig et al., 2013). Precision agriculture has progressed as a result of advancements in RS data collection, such as improved spatial and temporal resolution, spectral resolution, and a variety of sensor platforms (e.g., satellite, aerial, and ground-based) (Mulla, 2013). Precision agriculture recently demonstrated a significant increase in crop yield production (Loures et al., 2020; Singh et al., 2020). Spatial data mining techniques (e.g., hotspot detection) can be used with AgBD to identify crops produced in small geographic areas or a set of regions that are vulnerable to climate change and natural disasters (Jiang & Shekhar, 2017; Shekhar, Feiner & Aref, 2015; Vatsavai et al., 2012; Xie et al., 2017). Furthermore, consumer datasets and market manner can be used to improve food access and nutritional outcomes, and geo-social media can be used to detect and control food-borne illness outbreaks in real-time. AgBD could help agricultural decision-makers in four ways: descriptive, prescriptive, predictive, and proactive (Shekhar et al., 2017). The goal of the descriptive axis is to use AgBD data to characterize spatial and temporal variability in soil, land cover, crop, and weather characteristics, as well as to identify stressors, traits, and infectious disease risk factors that need to be better managed. The prescriptive way is to look for the required innovations for farm management. The predictive axis is a predictive analysis that uses historical datasets and integrated soil, crop, weather, and market models to forecast outcomes like crop yields and food insecurity. Predictive analytics can also be used to improve decision-making to forecast the spread of infectious agents and limit their impact on crops and livestock. Finally, the proactive axis includes crop development and stress observations from multiple farms across large regions and time scales.

The current state of AgBD can be divided into two categories: public and private data. Some examples of public AgBD are summarized in Table 2. Big data differs from one region to another based on various factors, including but not limited to the data availability, capacity building, and target of the study. Exploring sustainable FS necessitates detailed data covering all aspects of FS, putting pressure on decision-makers and scientists to initiate and prepare the necessary big data. To assist in filling this void, the big data paradigm should employ techniques, paradigms, and decision-making technologies, as illustrated in Fig. 2.

Methods and Models used for FS analysis

FS has four dimensions: access, availability, stability, and utilization; thus, incorporating such dimensions into agricultural models necessitates careful consideration and deep efforts. There have been few studies that have linked agricultural models with FS dimensions and indicators in order to understand evolving intertemporal dynamics and assess the effects of agricultural system intensification (Laborte et al., 2009; Laborte, Van Ittersum & Van den Berg, 2007; Marín-González et al., 2018; Morris, 2003; Nicholson et al., 2021a). However, such studies have focused only on a few FS indicators, such as household outcome, and ignored other dimensions and indicators. Therefore, we reviewed more than 1,200 related articles on FS modeling at the household and regional levels to assess the frequency of use of various FS indicators and make future recommendations to close this gap. Optimization models were used in FS but only on a few indicators of food availability (Amede & Delve, 2008). Crop simulation models are used to predict crop yield as an indicator of food availability, either as regression models (Beyene & Engida, 2016; Bharwani et al., 2005) or as complex biophysical models (Lázár et al., 2015). For food consumption expenditures, other sophisticated models, such as Holden & Shiferaw (2004) and Louhichi & Gomezy Paloma (2014) were used. Scopus database was screened for approximately 1,250 articles related to household FS models, and 130 articles were reviewed for which FS indicators were summarized (Table S1). We also looked for studies that looked at the determinants of dietary diversity at the individual level (most commonly, among young children or women) or at the household level. Dietary diversity, or the number of different foods or food groups in one’s diet, has been linked to several measures of household socioeconomic status that are frequently used as indicators of food insecurity (Jones et al., 2013). As a result, dietary diversity is frequently used as a proxy and stand-alone indicator of household food insecurity. We searched Google Scholar for relevant studies that provide empirical evidence about the determinants of the Household Dietary Diversity Score (HDDS), Household Food Insecurity Access Scale (HFIAS), and Food Insecurity Experience Scale (FIES). For this purpose, we used the following items in search: diet diversity determinants (130 articles), household FS determinants (870 articles), and experience scale of food insecurity (250 articles). From the 1,250 articles, 130 articles were reviewed. Figure 3 shows the network and associations of different models used for FS. It was found that statistical models were the most prevalent, and all models covered only two FS dimensions: availability and access. The detailed descriptions of these models, including type, classifications, the related references, and calculations of FS dimensions are presented in Table S1. Even though, many models and methods are used to investigate global FS, there are insufficient related studies in the GCC, particularly in the UAE, requiring much attention using the best tools to achieve most FS dimensions.