Enhancing Data-Driven Decision Making in Agriculture: A Big Data Approach

Increasing agricultural production to meet the needs of a growing population is one of the key challenges of the 21st century. Faced with significant population growth, shrinking farmland, and mounting evidence of the adverse environmental and health effects of agricultural inputs, we need new approaches to agricultural production that increase yields and efficiency on existing farmland. While centuries of agricultural research have greatly improved our understanding of agricultural systems, many factors that affect agricultural production remain poorly understood. Farmers often base critical crop management decisions on personal experience and intuition, because they lack quantitative scientific evidence about how those decisions impact yield. In order to meet the growing demand for agricultural products in an environmentally sustainable way, optimizing crop management decisions is becoming critically important.;A promising approach in the pursuit of providing farmers data-driven evidence to help them make optimal crop management decisions involves the use of large, historical datasets from commercial crop production. Farmers collect a great deal of detailed data about their farms and their crops as a byproduct of their everyday farming operations; here, we capitalize on this rich, existing data source, taking a "big data" approach to agricultural research. Considering cotton production in California as a case study, we amassed a historical dataset of more than 1,400 records of commercial cotton production. We mined this dataset to quantify how various factors impact yield and pest densities, in order to help farmers make better-informed, data-driven crop management decisions.;First, we quantified how both crop rotation and landscape composition---both factors that are very challenging to study experimentally---impact cotton yield and the density of a key cotton pest. Next, we expanded upon our agronomic analyses, using a combination of yield, pest, and financial data to quantify economically optimal management strategies of an important cotton pest. Finally, we mined the dataset using a variety of machine learning algorithms to develop predictive models of cotton yield and pest infestations.;Our results demonstrate the value in taking a big data approach to agricultural research. They show how datasets from commercial agriculture can be immensely valuable for deriving quantitative evidence about how factors impact yield and pest populations, and how to maximize yield and profits. Our results suggest that analysis of agricultural datasets using novel analytic approaches is an important complementary approach to experimental agricultural research, and will play a key role in generating the data-driven decision support tools farmers need to meet the ever-increasing demand for agricultural products.