Predicting Energy Crop Yield using Bayesian Networks

N.K. Newlands and L. Townley-Smith (Canada)


Agriculture, Bayesian Networks, Climate, Energy, Learning Algorithms


Natural resource problems typically must be modeled using data that is often incomplete, asynchronous and collected at different spatial and temporal scales with different levels of uncertainty. Variability due to climate, soil, pests and management decisions contribute to further structural and functional complexity of managed ecosystems. Bayesian networks are ideal for such situations by enabling diagnostic-reasoning on conditional dependencies to assess model structural as well as parameter uncertainty. We apply Bayesian networks to the problem of supplying regional biorefineries with an optimal, robust supply of biomass from energy crops. Crops have different optimal climate, water and nutrient requirements, and sensitivity to extreme weather, invasive pests and other impacts. Farmers adjust planting and harvesting times, available water and nutrients by irrigating and adding fertilizer. Increasing energy crop supply requires growing perennial grasses and fast-growing trees on less suitable agricultural land. We test a simplified model version (annual scale for a single crop barley-straw) in southern Manitoba, Western Canada. We examine the sensitivity of optimal yield to planting/harvest timing under historical weather, water, nutrients and extreme event/pest loss variability. We compare different classifiers in obtaining a network solution and discuss future work to apply the model at higher spatial and temporal resolution.

Important Links:

Go Back