Data science in agriculture

In the last issue, we introduced data science in the context of agriculture and highlighted its importance in precision agriculture. In this issue, we discuss some of the insights into the applications, advancements, and future of data science in agricultural systems.

Breeding and Screening of Crop Varieties

Data science is increasingly used in plant breeding. High-throughput phenotyping, which measures certain plant traits from the molecular to whole-plant level, has benefitted from the use of remote and proximal sensing to guide breeders through accurate trait selection (Singh et al., 2016). This type of non-destructive data collection has been adopted to crop species to identify and evaluate tolerant traits to abiotic (e.g., drought and flood) and biotic (pests, weeds, and diseases) stresses throughout different growth stages. Conventional phenotyping was particularly constrained, being a laborious, destructive, and time-consuming process (Beauchêne et al., 2019).

There have been several advancements in the types of sensors used to image, such as hyperspectral, fluorescence, thermal, and visible to infrared. These sensors can be used alone or as mounted on unmanned aerial vehicles (UAVs), robot-assisted imaging platforms, or cable-suspended camera systems to take real-time images and measurements throughout the selected time duration. Unmanned aerial vehicles are particularly useful for low-latitude, high-resolution aerial imaging to evaluate crop emergence, vigor, and yield potential of row crops (Sankaran et al., 2015; Shi et al., 2016). Moreover, big data and machine-learning techniques have been applied in the areas of molecular biology and biotechnology, including genomics, transcriptomics, proteomics, and systems biology (Ma et al., 2014) and to evaluate the yield performance of genetically modified crops (Johnson et al., 2019).

Crop Protection

There can be monetary and environmental costs that accompany pesticide applications currently required for pest management. While also costly, excess use of pesticides has been shown to increase the prevalence of pesticide resistance and negative effects on non-target species (Mallet, 1989). To improve pesticide recommendations, researchers have been using data science to accurately calculate the rate and timing of application of these agro-chemicals. This approach considers data collected about the pest, such as life cycle and biology, along with environmental factors such as natural predators, growth stages of the target plant, and weather conditions during the growing season (Meisner et al., 2016). Additionally, big data and machine-learning methods have become very useful to control weeds (Ip et al., 2018), detect foliar diseases (Ferentinos, 2018), classify field crop pests (Xie et al., 2018), and detect stored grain pests (Shen et al., 2018).

Remote Sensing

Remote sensing is another area where data science is intensively used. The common applications of remote sensing in agriculture are monitoring agricultural land use, estimating soil moisture, generating crop planting/intensity maps, classifying crops, assessing within- and between-field variabilities, forecasting crop yields, estimating evapotranspiration and irrigation requirements, monitoring crop health for input optimization, and evaluating ecosystem services (Weiss et al., 2020). Satellite missions such as Landsat and Sentinel have substantially improved their spatial, spectral, temporal, and radiometric resolutions in the past (Wulder et al., 2019). These high-resolution images are utilized to calculate various vegetative indices to evaluate the performances of different agricultural systems (Hatfield et al., 2019). For example, Xu et al. (2018) demonstrated how remote sensing can be used to estimate nitrogen uptake and cover crop biomass following winter, which could be useful in determining management practices for early spring.

Agricultural System Models and Decision Support Systems

Agricultural system models require a vast number of data inputs to simulate target outputs (e.g., crop development, water balance, nutrient dynamics, soil carbon, yield, net return, etc.) at farm and landscape scales (Antle et al., 2017). Moreover, advanced data science approaches such as deep learning have been used to train crop model inputs and outputs to evaluate the impacts of irrigation amount and timing on crop yield (Saravi et al., 2019). Wireless sensor networks (WSN), in combination with internet of things (IoT) and data analytics, aid in collection of various agro-environmental data, so these models can be integrated into decision support platforms. For example, Ellenburg et al. (2019) introduced Regional Hydrological Extremes Assessment System (RHEAS), an integrated modeling framework, to estimate onset, severity, recovery, and duration of regional droughts and expected crop yield outlooks. Moreover, decision support platforms based on web GIS are useful to connect local farmers to regional and global agriculture to improve demand-and-supply-based input provision, marketing, and agricultural policies (Delgado et al., 2019).

Combating Climate Variability and Change

Climate data enables researchers to evaluate the impacts of climate change on major crops and formulate and implement adaptation strategies such as climate-smart agriculture (Rosenzweig et al., 2014; Hassani et al., 2019). Data science together with the advancement in information and communication technology (ICT) is used in forecasting weather and seasonal climate, which could help farmers to deal with future climate variability (Klemm and McPherson, 2017). Improved understanding of the upcoming season would give producers time to make necessary changes to farm management. Long-term crop, soil, climate, and topographic data from large commercial farms have demonstrated the potential to design site-specific adaptations to climate variability (Martinez-Feria and Basso, 2020). Furthermore, artificial intelligence, Bayesian models, and neural networks have shown their potential to enhance climate warnings to extreme weather (Huntingford et al., 2019). Recently, Newlands et al. (2019) explored the applicability of deep learning to climate risk management options such as agricultural insurance. It was more accurate than the other approaches typically used and demonstrated its potential in lowering insurance coverage costs, among others.

Development of Smallholder Farms

Datasets that evaluate the production performance of major crops such as legumes (Cernay et al., 2016) and spatial databases like RiceAtlas (Laborte et al., 2017) can be utilized to enhance the food and nutritional security of smallholder farmers in developing countries. These datasets are particularly useful to identify the interaction effects of crop species × location × growing season × treatment combinations on yield, nitrogen content, and water use efficiency under different management practices (Cernay et al., 2016). Utilizing this information can allow farmers to predict issues surrounding environmental variability and to increase production efficiency across different agro-ecological regions. Data science also offers many opportunities such as increasing the effectiveness of agricultural development projects and generating better information for policy decisions. This involves creating systems that respond to local problems and can be utilized in future predicaments. However, it requires technological development in extension services, careful governance, and public investment in order to avoid a few growers monopolizing available space and reducing further development (van Etten et al., 2017).

Marketing and Supply Chain Management

Pre-season yield forecasting can assist on-farm decision making regarding product harvesting, storage, and marketing (Oliveira et al., 2018). Data science is widely used in tracking of food supply chains with the use of Radio Frequency Identification (RFID), intelligent supply–demand forecasting tools, and information management systems to ensure quantity, quality, and food system transparency while minimizing food losses along the supply chain (Tzounis et al., 2017).

Future Prospects

Data science has been identified as one of the breakthroughs to advance food and agricultural research by 2030 by the National Academies of Sciences, Engineering, and Medicine. However, challenges exist regarding quality, ownership, privacy, and security of data; integration of data types; data processing and analytics; and uncertainties in algorithms (Wolfert et al., 2017). Moreover, data science and scientific principles should be utilized together to explore data insightfully and to interpret the results accurately. This theory-guided data science will require interdisciplinary collaboration in research. With increasing open sources of big data, computational infrastructures, innovative methodologies, and public involvement in data collection such as citizen science (Ryan et al., 2019), data science has the potential to increase food security and environmental sustainability while promoting public- and private-sector initiatives and business ventures in future agriculture.