A stochastic planning framework for the discovery of complementary, agricultural systems

One of the greatest 21st century challenges is meeting the need to feed a growing world population which is expected to increase by about 35% by 2050. To meet this challenge, it is necessary to make major improvements on current food production and distribution systems capabilities, as well as to adapt these systems to expected trends such as climate change. Changing climate patterns may present opportunities for unidentified, geographical regions with adequate climate patterns to produce high-value agricultural products in a profitable and sustainable manner. This paper focuses on the design and planning aspects of a discovery process to unearth agri-food supply chains capable of generating attractive return on investments. A stochastic optimization framework is used to develop planting and harvesting schedules for a set of identified regions with complementary weather characteristics. To address the high-level of variability in the problem context, a two-stage stochastic decomposition method is used to consider a larger number of scenarios. As part of the solution process, a modeling scheme is developed that learns past interactions between entering discretized, weather scenarios and optimal first-stage solutions. In this context, machine learning and dimensionality reduction techniques are used to iteratively estimate each region's probability of belonging to first-stage solutions based on previous solution-scenario results. The implementation of the stochastic framework is shown through a case study applied to multiple locations within the US southwest states of Arizona and New Mexico.