Data assimilation to improve states estimation of a dynamic greenhouse tomatoes crop growth model
Mechanistic dynamic crop growth models have parameters values depending on the variety and culture method. For these reasons, the models must be calibrated under the specific conditions where they will be used. It is important to incorporate an adaptation technique for these parameters. The reduced simulation model TOMGRO was developed to predict the potential growing of greenhouse tomatoes. This model considers as state variables: the number of nodes, leaf area index and total dry matter. In this work, a data assimilation method based on the Unscented Kalman Filter (UKF) was developed, in order to improve the prediction on the three state variables of the TOMGRO model, by incorporating measurements from the crop. The equations of the model were solved numerically by using the fourth order Runge-Kutta integration method in the MATLAB-Simulink environment. The Kalman Filter was used to estimate the model states with different error levels. The filter performance was evaluated by the root mean squared error (RMSE) and also the Mean Absolute Error (MAE). The simulation results showed a better fit of the TOMGRO model when states are estimated using an UKF than when the model is calibrated by a standard procedure. Based on the statistic test results, it is concluded that the UKF successfully improves the prediction of the three state variables of the TOMGRO model. Therefore, the Unscented Kalman Filter is an efficient data assimilation method for non-linear dynamics crop growth models under greenhouses.
Torres-Monsivais, J.C., López-Cruz, I.L., Ruíz-García, A., Ramírez-Arias, J.A. and Peña-Moreno, R.D. 2017. Data assimilation to improve states estimation of a dynamic greenhouse tomatoes crop growth model. Acta Hort. (ISHS) 1170:433-440
unscented Kalman filter, error covariance matrix, sigma points