Estimation of Biomass Growth Using Unscented Kalman Filter

In the field of controlled-environment agriculture, a popular control objective is to maximize the yield of a crop, which is directly related to biomass growth. However, accurately predicting biomass growth is challenging, due to highly complex nonlinear dynamics, and the different time scales of plant and environment dynamics. Furthermore, some of the crop variables in the model needed to predict biomass growth are typically measured sparingly. To account for this last limitation, we present a state estimation approach that combines an Unscented Kalman Filter (UKF) and dynamic resizing of the measurement matrix. We present our simulation results using tomatoes growing inside a greenhouse for 250 days, with the UKF using the reduced Tomgro model. Using real measurements of biomass growth as a reference, we show that the proposed approaches can accurately estimate the biomass growth on a timescale useful for control (in minutes), using noisy environmental variables and very sparse (in weeks) measurements of the tomato plant variables. The proposed state estimation technique can be combined with full-state feedback control such as model predictive control to optimize the crop yield.