Confronting the Challenges of Big Data for Precision Agriculture

Big Data is challenging how we have been approaching agricultural research so far. Traditionally, agricultural experimentation has been based almost exclusively on the statistical principles elaborated by Sir Ronald Aylmer Fisher (1890–1962), which went a long way in helping researchers make sense of experimental results in the context of uncontrolled variation due to site- and time-specific effects. However, Fisher’s approach has always shown very limited predictive ability in situations outside the bounds of experimentation. Agricultural research is no stranger to the reproducibility problem that science in general is facing. Agricultural scientists have assigned more value to information about the quantifiable effects of factors and much less value to information about the farming environment that conditioned them. Agriculture is disorganized by nature. Uncontrolled variation tied to site- and time-specific effects has always undermined the transferability of agricultural science to the target users. The scientific methodology developed by Fisher has been followed so extensively that it is now largely defining the problems. As a result, farmers and scientists do not focus on the same kinds of uncertainties that condition the outcome of a management decision, and farmers have very nearly lost faith that the general scientific results produced by agricultural scientists can provide personalized solutions to on-farm problems. Farmers want to secure yield. They are risk averse in a world where soils are spatially variable and seasonal characteristics are largely unpredictable. The key to improving agricultural practices, and thus achieving productivity and environmental gains, lies in the personalization of recommendations for each production context. Big Data and precision agriculture will be the cornerstones of such evidence-based management in crop and livestock production. In crop-production agriculture, Big Data comes mostly from sensors on site, yield monitors, remote sensing by unmanned aerial systems (UAS) or satellites, and interpolated weather information (both cumulative and forecast). Currently, these sources of data do not talk to each other. Tapping into Big Data’s potential will require open data, standards, interoperability, and discoverability to converge, and that is currently far from the case. Precision agriculture tools, when coupled with innovative data-mining procedures and predictive models based on artificial intelligence, will be able to deliver personalized recommendations at an appropriate spatial scale, so that agricultural productivity and environmental performance can be truly improved.