Bayesian restoration of greenhouse desk temperature measurements

Modern greenhouse control solutions usually rely on temperature measurements collected in the greenhouse with high spatial resolution. The most important measurement locations are the close vicinity of the plants, namely on the plant supporting desks. Large number of desk sensors can ensure high spatial resolution, but increased volume of low-cost instrumentation results in a degraded reliability. Several methods have been elaborated in the past to handle the problem of missing or false measurements. This paper proposes a Bayesian approach for the substitution of missing data. The use of Bayesian networks has advantages as not only a single estimated value is returned but a distribution over the possible values as well. Furthermore the architecture of these networks is especially well suited for missing value replacement. The results show that the best overall performance may be obtained by the application of this Bayesian method, however in some other cases inserting the last known good value is also beneficial.