Dynamic Bayesian network for crop growth prediction in greenhouses

The paper presents an Internet-of-Things based agricultural decision support system for crop growth. A dynamic Bayesian network (DBN) relates indicative parameters of crop development to environmental control parameters via unobserved (hidden) Markov states. The expectation-maximization algorithm is used to track the states and to learn the parameters of the DBN. The steady state information is then used to derive a predictor for the measurement data a few days ahead. The proposed DBN avoids time-consuming training cultivation cycles, as only data of the current cultivation cycle are available to the algorithm. Three cultivation cycles of lettuce have been used to test the performance of the proposed DBN. The environmental parameters were temperature, solar irradiance and vapor-pressure deficit. The measurement data include evapotranspiration at granularity equal one day, and leaf-area index and dry weight, at granularity equal one week. It turned out that accurate measurement data prediction a few days ahead is possible even if the number of data samples is low.